WARNING: Summaries are generated by a large language model and may be inaccurate. We suggest that you use the synopsis, short and long summaries only as a loose guide to the topics discussed. The model may attribute to the speaker or other participants views that they do not in fact hold. It may also attribute to the speaker views expressed by other participants, or vice versa. The raw transcript (see the bottom of the page) is likely to be a more accurate representation of the seminar content, except for errors at the level of individual words during transcription.


Rethink X is a research organization dedicated to understanding the capabilities and impacts of new technology on our lives, economies and societies. They are exploring the possibilities opened up by automation and Artificial General Intelligence (AGI), and suggest a two by two matrix of barriers to entry and intelligence requirement to categorize occupations. Adam is taking a group to a virtual building, and Roblox is investing in infrastructure to make the Metaverse a reality. Ultimately, the Metaverse should be about experiences and identity, rather than possessions. With the right investments and resources, a quasi-utopia of automation could be possible in 10-15 years.

Short Summary

The speaker is discussing the implications of automation on intelligent labor and how it could disrupt many industries. They have proposed a two by two matrix to help think about the way automation might roll out, and are looking into AI tools that could help with research. The organization is focused on facilitating better navigation of the possibilities opened up by technology disruptions, with the mission to research, analyze and publish findings about the capabilities and impacts of new technology on our lives, economies and societies. The overarching goal is to have more thriving and well-being, and less suffering and destructive behavior.
Rethink X has increased its research and development from nearly 100% to 50% in order to disrupt foundational sectors like communications, energy, transportation, and food. Adam has called for people to deposit money in a crypto exchange to speed up the process. Sam Bankman Freed is advocating for resources to solve AI Safety Research. Elites are moving to shorter timelines and the public is mostly at long. Examples of new technology, such as self-driving cars, generative art and personal assistants, can make automation technology feel more real. When people realise they have lost their job to a machine, this could be a transformative moment.
The transcript discusses the concept of automation and its implications for jobs. The idea is to identify which jobs are most vulnerable to automation and which people should prepare for being technologically unemployed. It is proposed that occupations can be categorised into four quadrants based on barriers to entry and intelligence requirement. As an example, the automation of driving is discussed, due to the low barrier to entry and high stakes. The concept of cognitive labor that is low skilled from the perspective of today's systems is also discussed, such as cashiering, retail, janitorial and custodial services, and landscaping. The mistake of our civilization is that we judge people based on the work they do.
The speaker discusses the idea of creating a meritocratic hierarchy around occupations, and the potential problems that may arise when artificial general intelligence emerges. They suggest that the risk and reward of professions such as scholars, poets, writers, musicians, and artists is not high, and that it is important to consider how likely certain jobs are to be automated when choosing a career. Programming and software engineering are examples of occupations that may be less likely to be automated, whereas radiology and teaching may be more susceptible. Technology is making teaching and tutoring more automated, but personal relationships may be too difficult for AI to supplant. Landscaping is an example of an occupation that involves time and solitude, and can still be enjoyable.
The speaker discussed the impact of automation and Artificial General Intelligence (AGI) on the future of work, claiming that in 10-15 years, there could be a quasi-utopia as a result. They suggested that the conversation should focus on how to organize society during the transition period, given the limited time frame. They also pointed to investing in renewable energy as an opportunity to transform the Australian economy, creating many jobs in building out renewable energy infrastructure, engineering, coordination, management and designing markets. Finally, they highlighted the increasing automation of coding tasks, with AI models becoming increasingly sophisticated and capable of taking on mundane and repetitive tasks.
Adam is taking a group to a building they recently bought from a Robloxian coral rhyme. He explains that the building is more demanding than the peak, but game streaming should keep the performance at the same level. They are interested in exploring indoor environments and using them for different purposes, such as virtual architecture, which could be a cool job with endless possibilities. Adam believes that imagination is the limit with virtual architecture and it could be a great job in the future.
The transcript discusses the difficulty of building a metaverse and the need for immense amounts of hardware and data centers to support it. Mark Zuckerberg and Tesla have invested heavily in AI related fields, and venture capitalists are likely betting on the development of AGI, even if they don't necessarily believe it will happen soon. Roblox is an online platform which allows users to create and play games, and is investing in infrastructure to make its voice chat work at scale. Ultimately, the Metaverse should be about what experiences can be had and who people can be, rather than what can be owned.

Long Summary

The previous session discussed two groups of people - Tech leaders and Elites - and their response to AGI timelines. Last week, the discussion focused on the motivations of mathematicians, and how their work is incentivised. The prospect of machines being able to do the parts of mathematics that mathematicians currently do has changed the way some mathematicians view their work and their career prospects. The discussion explored why people continue to do mathematics despite the opportunity cost, and the trade-offs they make to do so.
Adam has been reflecting on the changes in machine intelligence and artificial intelligence over the past few months and how it affects his career trajectory and the research his team and organization does. He is considering how the changing timeline affects their mission and how it should affect his own career trajectory. He is also considering how it affects the pursuit of intrinsic beauty and how it compares to spending time with family and other things that are valued.
The speaker is discussing the potential for automation to disrupt many industries, including intelligent labor. He proposes a two by two matrix to help think about the way automation might roll out. He is also looking into AI tools that could help with research, and expects to be onboarding them in the next 12 months. This disruption of intelligent labor is seen as categorically different to other forms of disruption, and could have a big impact on the global economy and social life.
The speaker is discussing how AI research assistant software is likely to impact their workflow in the next couple of years. They mention Facebook's new Galactica tool, which they tried out on mathematics and found to be reasonably accurate. The speaker is excited by how this technology could make them smarter, but is aware of the need to keep up with the fast-moving field. The mission of their organization is to research, analyze and publish findings about the capabilities and impacts of new technology on our lives, economies and societies.
The organization is focused on facilitating better navigation of the expanding space of possibilities opened up by technology disruptions. This allows us to make better decisions about our future and maximize our values. There is both a need to figure things out and to change minds when it comes to navigating this space. AI tools and progress will have an impact on the mission, but it is not a linear process. It is important to keep in mind that the overarching goal is to have more thriving and well-being, and less suffering and destructive behavior.
The percentage of research and development for Rethink X increased from close to 100% in 2017 to at least 50% in the present. The mission of the organisation is to change the Overton window and to get people to think and talk differently about what is possible for the future. This mission has a finite life, and the disruptions don't take forever. The biggest disruptions are in the foundational sectors, such as information and communications, which were massively disrupted by the internet. The goal is to get people to be more optimistic about the future, and the project is likely to end in the next 10 to 15 years.
Communications, energy, transportation and food are all foundational sectors currently undergoing disruption. These sectors are mutually capitalizing and accelerating one another, and with the addition of automation, the disruptions are likely to proceed even faster than expected. This puts a time pressure on generating and disseminating insights quickly to make better decisions before it's too late.
Adam has experienced a sense of urgency around producing insights that can be useful in the context of AI/AGI acceleration. He suggests that people should deposit money into a crypto exchange and gamble it in order to get resources to solve problems quickly. The group discussed public timelines and how FTX, a crypto exchange, is likely part of the mix in terms of people's sense of urgency. Insiders are starting to move to shorter timelines, but it is important to not read too much into it.
Sam Bankman Freed, the founder of FTX, has been advocating for the gathering of resources to solve existential risks such as AI Safety Research. This event of tens of billions of dollars disappearing is partly due to the urgency to act fast. The world is expected to get weirder in the next 10 years as people make bets on the future and the resources to be gained from AGI. Elites are moving from long to medium timelines with a median of 2040, while the public is mostly at long with a vague unease about AI taking over.
The speaker discusses how getting hands-on experience with new technology can open up possibilities and how this was seen with the first iPhone. They then suggest several other examples, such as self-driving cars, generative art and personal assistant, which could make automation technology feel more real. They add a more pessimistic thought, that when people realise they have lost their job to a machine, this will be a transformative moment. They suggest this could happen when people who drive for a living, such as truck drivers in the US, start to see self-driving cars and then lose their job to a machine, which will be a Flashpoint.
The prediction is that the loss of jobs to machines will cause a rapid shift in public sentiment. It is likely that conversational AI, generative art, music, and video will be commercialised first, before self-driving cars reach a high level of reliability. It is similar to the shift in public perception around internet dating, which was initially seen as something only for 'weirdos'. It demonstrates how quickly public opinion can change.
In the early 1990s, it was considered crazy to buy anything on the internet or to date someone over the internet. The speaker then proposed a schema to think about occupations that are threatened by automation, which consists of two axes: the barriers to entry of the occupation and the intelligence requirement. The example of drivers was used to demonstrate that driving a taxi is a much harder task than many other tasks that have already been automated. The speaker warned to be careful about how the term 'intelligence' is used in this context.
The transcript discusses a concept of barriers to entry and educational attainment requirements as two axes in a graph. Low and high stakes of an occupation are represented on the vertical axis, while the barriers to entry are represented on the horizontal axis. This concept is used to illustrate how occupations can be categorized into four quadrants. Automation of driving is used as an example, as the barrier to entry is low, but the stakes are high. This means that there is an incentive to automate the occupation quickly, as the reward is great and it scales.
The original idea was to identify jobs that were most vulnerable to automation, and which people should prepare for being technologically unemployed. However, it became clear that this intuition was likely to be false. To stimulate conversation, it is important to distinguish between low skill and physical labor, as the latter may require more cognitive ability and training than originally thought. This term is complex, and the observation is that people may need more training than initially thought to perform the job.
The transcript discusses the concept of cognitive labor that is low skilled from the perspective of today's systems. It also looks at low stakes and low barrier to entry jobs that humans can do, such as cashiering, retail, janitorial and custodial services, and landscaping. It is noted that these are junior positions that most humans can do and that nothing said is meant to look down on people in any way. The mistake of our civilization is that we judge people based on the work they do.
The speaker is discussing the idea of constructing a meritocratic hierarchy around occupations and how this could become a practical problem when artificial general intelligence emerges. They suggest that all humans will be rendered useless in comparison, and that it would be beneficial to get ahead of the curve. The speaker then mentions that there are jobs which require more training and qualifications, but the risk and reward are not necessarily high. They mention that this includes professions such as scholars, poets, writers, musicians, and artists, where the stakes of life and limb are not at risk.
The transcript discusses the concept of a chart that divides occupations into four quadrants based on the barrier to entry and stakes associated with them. The top left quadrant includes occupations such as driving and food preparation, where the barrier to entry is low but the stakes are high, and people are often underpaid for the risks they take. The top right quadrant includes professions such as surgeons and lawyers, where the barrier to entry is high and the stakes are also high. People in these occupations are typically better compensated for the risks they take.
The speaker talks about a schema of professions divided into four quadrants based on the likelihood of automation and the reward or risk from the employer's perspective. Over the last decade, some jobs have fallen to automation more quickly than expected. There is not an equal pressure on all occupations, as those with high stakes are more likely to be automated. The speaker suggests that when choosing a career, it is important to consider which jobs are more likely to remain safe for longer, as well as the potential for new markets to rapidly grow in size.
Programming and software engineering have been around for decades, and despite the large increase in the number of people doing computer science, the wage premium for these jobs has remained. This is likely due to software eating the world, as more applications are found for programming. Radiology is an example of a field that may be subject to automation in the future, whereas teaching may not be as easily automated. AI tutors are being trialled and AI bots may soon be able to stand in for teachers. This makes it difficult to predict the demand for certain jobs, and the wage premium associated with them.
Technology is making teaching and tutoring more automated, but there may still be more demand for teachers. Occupations that involve personal relationships, such as teaching, coaching, and therapy, may be more difficult for AI to supplant. It may be possible to rehumanize occupations that have been under pressure to dehumanize, such as teaching, and spend more time and care with each other. Landscaping is an example of an occupation that involves time and solitude, and can be enjoyable.
Gardening is an example of a task which can be automated, but the process of doing it and the experience of enjoying the result is not the same. There is a potential shift in humanity's attitude towards work due to automation, which may lead to more care and investment in certain tasks. This is already being seen in the disruption of education, with more people homeschooling and using online tutors and AI tools to facilitate this. This is allowing parents to spend more time with their children than they would otherwise.
The coming disruptions from remote work and AI tools will allow people to solve coordination problems in ways that were not possible before, leading to the creation of new occupations. Online tutoring is one example of this, and it has become more popular as people live differently. The danger of this conversation is the arrival fallacy, where people think that the future will be static. AGI may also lead to a change in jobs, and how we respond to this change is an important consideration.
The speaker suggests that the transition to a world with AGI and the singularity may only take 10-15 years. This means that the conversation surrounding this transition needs to be reframed to consider the limited time period in which it will take place. During this period, there will be an interim period where people will still need to work, but eventually there could be a quasi-utopia on the other side. The speaker suggests that the conversation should be focused on how to organize society during this transition period, as it will only last for a few generations.
The speaker talks about the chaotic and difficult interim period that is coming in the next five years, and how one should make career decisions accordingly in order to avoid ending up unemployed. They then move on to discussing the idea of transforming the Australian economy by investing in renewable energy. This will create a high demand for energy and raw inputs, and thus many jobs in building out renewable energy infrastructure, engineering, coordination, management, and designing markets. This will ensure that these jobs remain safe and highly valued for the foreseeable future.
The demand for energy will continue to increase due to the use of AI in data centers. Coding is becoming increasingly automated, with Google citing their internal GitHub codex model as increasing programmer productivity by 6%. This is impressive considering the high quality of software engineers at Google. Although AI models currently exist, they are flawed and make many mistakes. However, it is likely they will become much better. Most programming work is mundane, repetitive and low cognitive labor.
Humans and AI are increasingly being used together to automate software engineering tasks. This has the potential to unlock opportunities on a much larger scale than before. However, it is uncertain whether this is a stable equilibrium, as was seen in the example of chess. It is suggested that when making predictions about other occupations, it is useful to think of them as tasks that can be broken down into subtasks.
There are many occupations where having a second human being present is beneficial. In the future, AI-enabled general purpose robots could help people in these roles, providing an extra pair of hands or even just standing by with a wrench. This could be useful for occupations such as plumbing, and even for hobbies. This is not something that will happen in the next five years, but in the next 15 years it is possible that machines will be used in an additive way, rather than a replacement way.
The task of implementing a new product or feature at a company like Google involves a lot of process and coordination, not just coding. AI is now being used to automate low and medium level coding, and startups are trying to exploit these new capabilities to do new and interesting things. On the other hand, when it comes to home improvement projects, there is a design and engineering dimension to it, and then the implementation of the design requires specific expertise such as carpenters, builders, plumbers, and electricians. People who are handy or masochistic can also do the work themselves.
In this discussion, the participants discussed how AI can be used to design and coordinate certain tasks, such as designing a bathroom or house. They discussed how it would be much cheaper and easier to get the AI to write code that's mostly been written before, and come up with ideas that interpolate between things that already exist. They also discussed that it would take longer for the AI to generate a reasonable solution to problems that don't have a lot of precedent. Ultimately, the participants concluded that AI can be used for design and coordination, but it may take some time for it to be able to generate new ideas.
Adam is taking everyone to a building they bought from a Robloxian coral rhyme. He explains that it is more demanding than the peak, but the performance should remain the same due to game streaming. He shows them a lecture hall and talks about how they are interested in exploring indoor environments and using them for different purposes. He also talks about the future of virtual architecture and how it could be a cool job, as the sky is the limit with imagination.
Imagination is the limit when it comes to creating experiences in virtual worlds. This has been a popular topic of discussion for many years, with references to Star Trek, The Matrix, and more recently, NFTs and the Metaverse. It is possible to unlock creativity and grow the space of possible experiences, however the focus has shifted to what can be owned and possessed, which is something many people find discouraging and dystopian. Ultimately, the Metaverse should be about what experiences can be had, and who people can be, rather than what can be owned.
Roblox is an online platform which allows users to create and play games, and is currently being used to create a virtual world. The platform has its own monetization system, however it is not intrusive and does not require users to pay to join. Roblox is investing in infrastructure to make its voice chat work at scale, however it is difficult for open source to compete with this. There are issues with using Roblox in certain countries and for certain people, and open source needs to find a way to collect, coordinate and deploy capital in order to compete with the new trends in technology.
The transcript discusses the difficulty of building a metaverse and the need for immense amounts of hardware and data centers to support it. It also mentions the need for a country to coordinate and deploy its underlying fiber in a community fashion to have a truly open metaverse. The discussion then shifts to cryptocurrency and the potential for it to coordinate large systems of people doing good things, but there is still not much evidence of this being done in good faith. Finally, the speaker suggests that it may still make sense to invest heavily even if there is only a small chance of AI happening before 2030.
Mark Zuckerberg has invested heavily in the metaverse, but it is really an investment in AI, as it will involve a large AI component. Tesla is another large bet on AI, and the level of investment in AI startups and related fields has skyrocketed in the last year. The state of AI report by Ian Hogarth and others details the level of investment in AI and other fields, from pharmaceuticals to synthetic biology. AI and related investments are being made in many different areas, and the bets are becoming increasingly risky.
Venture capitalists are likely betting on the development of AGI, even if they don't necessarily believe it will happen soon. Bayesian probability allows betting on multiple candidates, as the potential payoff outweighs the risk of loss. It is hard to lose if you spread the investment across a reasonable number of candidates, although clandestine government projects cannot be bet on. The conversation concluded with the reminder to continue the discussion next week.

Raw Transcript

uh Matt asked for a recap of last time uh so maybe I'll briefly do that so last session uh Adam wasn't there and we went off in a bit of a tangent uh they were two weeks ago we were discussing what you see on the boards uh in front of you so we were talking about AGI timelines short medium and long and the responses of different groups of people to events that might shift them from one believing in one timeline to believing in another we've kind of spent a fair bit of time talking about two of the groups the two groups on the second set of boards that you see are Tech leaders or insiders so that's researchers engineers investors and the second group were Elites so a cultural political and economic Elites and we discussed you know in a fair bit of detail how those first two groups what we maybe think their current beliefs are and what the shifts have been and what they may be as various things start to happen and we just started talking about the response of the public where they are now and what might change people's minds that was two weeks ago so there's some threads we'll pick up there uh last week very much in the vein of I suppose the public view on AGI timeline shifting I was one way into that is just to speak honestly about my internal thought processes around it and how it's affected my own view of my career and what I do and uh that's you know I think uh I've been thinking pretty hard about this for probably five or six years now because I expected to be more or less where we are now and I have expectations about what's going to happen over the next 10 years as well and there's that was around you know why do we do mathematics what's our motivation for doing it what gets us up in the morning and more importantly what keeps us doing it even though we're sacrificing other things we value in order to do it so a lot of the difficult cognitive labor that people do it's incentivized in various ways I mean outside of Academia you know there's a profit motive and various other motives inside Academia you know uh you're basically paying a significant opportunity cost to be in Academia if you're intelligent enough to have a position as a permanent as a mathematician you can probably go and do things that earn multiples of the salary you'd get in Academia with various other trade-offs but uh you could make more money and have a easier life to some extent so why do it um so I was discussing my thoughts around that and how that's affected by the prospect of machines being able to do the parts
of my job that actually are the things that Moto and me do it within some reasonable time frame but not saying mathematicians research level mathematicians will necessarily be replaced in five years or something but I don't think it's 50 years away so that should make a difference if you're thinking about the the contribution you make towards a field as being kind of cashed out over a very long time Horizon if you care at all about it being cashed out and it's not just some pursuit of intrinsic Beauty And even if it is a pursuit of intrinsic Beauty how do you feel about the situation where you're really playing in the Kids Corner at some point is it still worth you know uh spending time doing that basically to satisfy your hedonistic impulses to just do pretty stuff when you could be spending time with your family or something else that you also value so that was kind of the topic last time um maybe that's a good place to jump off I don't know if you got a chance to to watch it Adam but what I just said probably gives you enough to go by uh how is your updated timelines changed your motivational systems and how you value what you do and uh how you think about adapting to those changes in your beliefs yeah I've been thinking about this a lot just I wouldn't say over the last five years but but certainly over the last four to six months I think I've been taking the possibility that that of a a sharply accelerating trajectory for machine intelligence artificial intelligence and the ways that we've described it um up to and including the possibility of fully sentient self-aware Sapient artificial general intelligence I I that we seem to be on a steeper trajectory that I would have imagined if you if if I think back to my earlier self three five years ago so on and I was fairly bullish back then I mean following the space for 20 plus years but I suppose the Transformer and and diffusion approaches have made very substantial leaps in progress to not just in in raw capability but towards generality that I wasn't expecting um and so yeah I've been thinking about over the past four to six months how should this affect you my own career trajectory but perhaps more interestingly for this discussion how does it affect the research that my team does that my organization does how does it what what does that changing timeline do to our mission for example um so I can speak a little bit about that that's the first thing I thought I would I would um open with and then uh don't let me forget there is I I
many years ago I proposed a sort of for fun a blog post long ago I don't know six or eight years ago a schema for thinking about um uh for one one way to kind of think about how um certain careers might be automated like in what order and what what no not order but but uh how might we think about the ways in which automation would roll out given uh competing pressures and priorities and so there's a there's a two by two Matrix that um uh I offered for that maybe we could put that up and expand on it a little bit later so that's if we have time we can get to that part of the um of the thinking that I thought maybe I would I would share today to start on my own um personal thinking and then that of my organization which is an independent think tank of course and we we we're trying to understand the significance of technological change and the the choices that that societies um now face as a as uh we are looking out into a future where we need to fully expect abrupt and impactful transformation of substantial portions of the global economy and and some significant elements of social life as well as a consequence of technological advancement um I mean so so that's that's disruption basically in a nutshell and it's one thing to talk about energy Transportation food Health Care um education uh those sorts of domains and any narrower industry with that falls under those sectors it's one thing to talk about disruption of those things that's sort of or you know in in some sense shallower um but the disruption of of intelligence and intelligent labor um which is which is an input into virtually everything uh is isn't It's seems to me to be categorically different so I'm again I'm still trying to figure out how this affects us for me personally um it hasn't changed the way that I do research yet but that feels like it's coming so I think that that we are over the probably over the course of the next 12 months my team will start investing considerable time and energy looking at new AI tools that are available that will actually help us do our work I know I'm aware of several things that are in early stages and the possibility that I see is that these can mature very quickly um so the idea that you could have an AI research assistant for example that can um uh read and digest literature very quickly do fairly smart things about data Gathering and simple analysis and so forth that these these I think I think there's it would be naive to expect to not be onboarding those tools over the next
within the next couple of years in a field like mine uh if you want to stay relevant and at The Cutting Edge I just I just I think that that's almost a necessity I got I suppose the alternative would be if you had um a very very large team you know lots of lots of junior staff you can imagine sort of a a operation with lots of internets for example and you could throw smart people at your problems like that well that would be a similar situation but my organization isn't in that situation um and so we would sort of what what are we likely to do higher uh 10 Junior researchers or just you know licensed the AI research assistant software and everybody on the team starts using it I'm guessing it's going to be the latter so that's the first thing that's that's in Practical terms I think this technology is going to impact my workflow did you see Galactic ideally sorry good did you see this model really did you see the news yesterday about Facebook and papers with code they uh they have a new thing they released called Galactica which is a science assistant trained on some 40 million papers from across science and other areas and you can ask it to like write a blog post about X and it will do a reasonably good job I tried it on an era of mathematics and it looked pretty legit to me so yeah there's uh the descendants of that Tool uh definitely capable of doing some of the things you're describing maybe even that tool already yeah yeah exactly um I I haven't surveyed that space myself yet so and as as you I think you said it was yesterday so this is a space that's I mean these things are happening Fast and Furious it seems like every week something new is coming along um so yes that that these new capabilities will definitely be impacting my workflow uh and just personal process as a researcher going forward in the near future this is not something that's five or ten years away but this is sort of imminent I feel um and I'm kind of excited because I'm hoping it will make me smarter effectively but uh uh you know we'll we'll have to see um as a team in an organization the or rather I suppose I should probably zoom out to the level of mission um the mission of my organization is pretty straightforward it's it's to uh you know conduct research and and perform analyzes develop and publish and widely disseminate findings um about uh these new technology uh and its capabilities and the likely impacts those are going to have on our lives on our economies on our societies
and so forth on Indus specific Industries right down to individual sort of markets um and uh with forewarned being forearmed with that knowledge uh allowing us all to make better decisions about our future where to invest where to divest and so forth um and uh I think the the overarching picture is that if a new and much larger space of possibilities is opened up by uh by these technology disruptions as occurs with any new tool of any kind any time new a new tool um anytime a new technology anytime new knowledge and I Define technology I think I've defined it in here before as practical knowledge I find I Define technology as a form of knowledge um any knowledge or tool expands your degrees of freedom right it it opens possible Futures to you uh that were foreclosed before and that spot possibility space expands and the question is how do you navigate that and and you know uh and and why how and why do you navigate it or maybe why is not the right question is it's not the right way to look at it but but to what end so given given uh a set of goals or values that you're trying to realize or optimize or maximize or move towards um whatever they might be uh the the as a new space of possibilities opens the possible ways to navigate through that space to the end of of maximizing or realizing whatever your set of values it might have that is the uh that's that's what my organization is aiming at doing is is facilitating better navigation so that uh we have we have more more thriving and well-being um and less suffering and uh destructive behavior and so forth on that note along the way on the journey so here's a question and sorry go ahead yeah so you rethink eggs What proportion is figuring stuff out and What proportion is changing Minds in terms of their relative importance I mean you do both and maybe it's in the process of Shifting I infer putting more weight on the latter but if you had to put a numerical figure to it what percentage I mean you could have organism I mean it could be useful to basically be 1090 right like maybe there isn't so much stuff to figure out but there's lots of Minds that should be changed or it could be the other way around I'm curious because that seems to have some bearing on how AI tools and progress will impact the the mission sure sure yeah let me let me speak to that very quickly and then I'll return and complete the thought that I had a second ago because it ties together with this so to answer your question um I don't think it's linear but it's
the the the percentage went from uh close to 100 percent just about the research and developing the insights in year one which was 2017 for my for rethink X and it's probably closer to I don't think we're at 10 90 at 10 90 now but it's probably closer we're certainly past 50 50 at this point where where and my role is is I mean I suppose I'd probably become I've been I've been you're a thought leader Adam that's where you've been no I've been promoted to my level I've thrown into my level of income promoted up upwards in into a management position where I I now direct the research team I don't do as much of the research myself and one of the reasons why is because it's now uh one of the responsibilities that I now have is in the the dissemination um domain where we're trying to to broadcast the information and the insights that we've generated and uh again the goal of the organization is not to is not to sort of have all the answers and tell people and give people marching orders that's not that's not really it the the goal of the organization is to change the Overton window uh to change the conversation about what is and isn't possible and and if our mission has succeeded if we've gotten people to think and talk differently about what's what's what is possible for their future um and there's a there's a valence to that the valence is that the with all these amazing possibilities some of them are astonishingly beautiful and positive and bright and in ways that many people with their existing window and view to the Future um don't see and are unduly pessimistic and so forth um now the way this wraps around who AGI in particular is that this mission um has a finite life there's a time frame on this it's fairly short right I mean these if we are very busy telling everybody well these disruptions happen quickly they don't take 50 years they only take 10 to 15 years and now I'm five years into this this project as it were well um I mean I'm not I'm not going to be doing this forever um because the disruptions don't take forever and that I've been screaming that from the rooftop so so there's a there is a a uh a finitude um to this entire project and to the mission now you could say well they'll just be more disruptions following on afterwards and we could carry on afterwards but um okay maybe but the the big the big big ones um are these what we call foundational sectors and the internet disrupted information and Communications in a massive way the information and
Communications are foundational sectors you can think of them as one thing or separate depending on how you want to look at it but at any rate the internet in conversions with of course various Computing Technologies um uh disrupted that foundational sector first led by 20 or 30 years uh now but pretty historically pretty much right on his heels we have three more foundational sect four really uh foundational sectors coming three it fit into our schema energy transportation and food okay all simultaneously pretty much right now energy a little bit sooner food a little bit later but but basically all in another 20-year period or so and now then on top of that you then have Labor if you want to think of that as a sector which may maybe it's not that's not the right way to think about it but we have Labor potentially being disrupted by Ai and automation okay well one of the things that we are arguing is that under these exceptional circumstances where these where these major foundational disruptions are all happening simultaneously one of our arguments is that they are they are mutually capitalizing and accelerating so they are each of them individually is very likely going to to proceed to unfold faster than uh in the context of the others than it would on its own other things being equal um and so we I mean we haven't Quantified that but that's just that that is a theoretical prediction uh that is remains qualitative we don't have a way to quantify that but we we uh would say something along the lines of other things being equal expect these disruptions to proceed even faster than your intuition might expect because they are all accelerating one another energy clean energy accelerates clean Transportation clean Transportation accelerates clean food and so on and so forth okay well throw a automation on to all of that and it's like having a barbecue and then just just you know backing up with a fuel tanker and just dumping gasoline on it it just it I mean I I can't imagine this is going to do anything except accelerate all of this um how could it not and so uh this is all very I get sorry a too long-winded way of saying that I now feel even more time pressure to get insights generated fast enough and then get them disseminated fast enough and widely enough to actually make a difference because uh you know before it's too late and by too late I mean um it's too late to make to make use of in in making better decisions as opposed to worse decisions uh in the context of
of all of the transformation that we're experiencing so that so if this is this is the main thing I suppose for right now immediately right now um that the the AI AGI uh acceleration has done for me it has it has created a massive sense of urgency around all of the other work um and it wasn't like I didn't feel any urgency at all before but it now I feel like oh my gosh if I we have to any insights that that we are capable of producing we better produce them really quickly like within the next several years or they're just not going to be useful um so that is the that's that's the context in my in my own personal work and in the context of my organization hmm yeah thanks uh do any of you in the uh audience have any comments on this topic maybe taking off from last week's uh seminar if you were there or you watched it or what Adam just said interested in other texts scary yeah well obviously the solution Adam is that you should get a lot of people to deposit money into your crypto exchange and then gamble it in order to get resources to solve the problems quickly enough that's right I just have to just have to tell everybody all 16 billion are going to the effective altruism movement and uh it'll all be fine so get on it just don't just don't give it away too early you might have to choose a different strategy now but I'm sure you can come up with something um yeah all right uh yeah I don't think I have anything particular more to say in that direction so maybe let's pivot back to some of the threads that were left hanging two weeks ago if anybody has further comments on that I'm interested to hear them but uh okay so over here on this board uh I guess we started talking about public timelines and I suppose that's uh in some sense what we've just been discussing with I don't know if we're insiders or certainly not Elites um okay so we talked about uh well maybe let's recap where we kind of thought the other groups were at so on the fourth board here there was i l goes to M to S with a question mark this is the Insiders who are kind of at least a influential subset uh starting to move to Shorter timelines I think it's probably too simplistic to see what happened with FTX as being too much about short AGI timelines but I think it's almost certainly part of the mix in terms of people's sense of urgency I think you could read too much into that and I think some people have so I wouldn't want to put too much weight on that but I do think that uh our FTX um so the the crypto exchange FTX just
collapsed last week or uh over the the last week uh so that includes the the future fund and many philanthropic things Downstream of FTX uh and the founder of FTX uh Sam bankman freed has been going around talking about how it's important to gather lots of resources in order to solve existential risks and one of those is you know that they funded is a AI Safety Research so there's uh yeah I don't know what weight to give it but it's still quite surreal to be sitting here talking about how maybe like a a financial event of the order of magnitude of tens of billions of dollars disappearing is partly the malfeasance that's going on there uh to the degree it is malfeasance and not just incompetence is at least partly a result of needing to go fast and that speed is to do with the same kind of urgency that Adam is feeling so okay again maybe don't take that with a grain of salt with this particular event but I do think it's appropriate to see this strangeness as a kind of marker of future strangeness so one of the ways in which you'd expect the world to get weird over the next 10 years is that if people have wide widely diverging expectations about what the future looks like and some of them think there are trillions or tens of trillions or hundreds of trillions of dollars to be allocated with immense productivity gains and people will stop making bits based on that and if you think bubbles based on past expectations like the internet boom the possible resources to be freed up and productivity to be gained from AGI once it starts looking plausible or more plausible than it is already the bets that people are going to make on that will look all past bubbles look like a joke so FTX will I think not seem so big compared to some of the stuff were going to happen and that's we're going to see in the next five years it would be my prediction so expect more of this um okay so on uh yeah so we talked about Insider timelines and maybe some of them are moving to short timelines um Elites maybe are starting to move from long to medium meaning a long meaning sometime you know in a century and medium uh how did we Define medium medium was having a median uh of um of 2040. okay and then the public uh yeah I don't have a good sense really of where the public is I guess uh I think probably mostly at long but with punctuated by kind of being aware like a kind of vague unease about things like stable diffusion and I mean everybody knows about sci-fi storylines about AI is taking over and
doing nasty things and so on but I feel like it's more or less completely divorced from expectations of daily life I don't know I'm interested in your opinions on that what do you I mean what is the general normal view on to what extent this is going to change life over the next 20 years well I I have one thought on this it builds on something we talked about a couple of weeks ago uh which is um a couple of weeks ago we talked about how getting a hands-on experience with the new technology can awaken you to possibilities and we mentioned a few examples maybe some that we've had in our own lives I certainly remember the first time I used an iPhone and that was transformative it just instantaneously felt like oh this is this is so this is what the future of handheld Computing and Communications is this thing the smartphone um even though it wasn't perfect and and and it took you know a number of years before it it realized its potential functionality it was that was a transformative moment um I suggested several other examples one would be when you know the first time a person sits in a car that drives itself that is going to make uh automation technology feel very real um in in concrete in a way that the abstraction doesn't feel real uh even if even if even if it it is real in evidence around a person um without that immediate direct firsthand the knowledge and experience they won't grock it they won't they won't really internalize the reality of it I think I think so we talked about that I would like to add to the list of things we already um the large other examples self-driving cars um generative art and personal assistant um I would like to add and this is maybe a little bit more pessimistic and dystopian um when people begin to realize they have lost their job to a machine then that will be a that will be an a remarkable waking up moment for anybody that that happens to you and I think that it won't have to happen to very many people before that just it becomes an explosive wave of uh transformation of public consciousness um now I think these things could happen together for example if if you drive an Uber or a truck for a living in the United States or elsewhere but in the u United States uh we have a lot of people who drive for a living especially because of trucking long distance Trucking um when those people start seeing self-driving cars and they then lose their job to a machine that will be I think um a kind of pivotal or um uh you know Flashpoint perhaps so that's another
prediction of mine is that is that uh in individual people's lives loss of their their jobs to machines will act like a flash point and um uh I think public sentiment could shift could swing pretty rapidly um if we have even a you know a relatively small number of those examples I don't know what the those numbers would be like what fraction of all jobs or what fraction of population but it's probably a very small number um before the the uh fear drives a massive shift in Consciousness towards fear and then uh either expectation or concern that we we're we're in a short scenario the S scenario as opposed to the L of the M scenario [Music] foreign yeah maybe I'll add another one here as well which is um you had to pick up something you said earlier about expected timelines for people losing jobs or what things will be automated first I think it's pretty clear that a lot of including my expectations were more or less completely backwards so for a long time people expected things like mathematics or maybe even driving mechanical operations to be automated first but partly because there's just so much data of humans talking and generating images and and doing other stuff on the internet and we're much better at feeding text through these large models than anything else it looks more likely that conversational AI or generative art music and video these kinds of things will be commercialized first and reach a level of Polish first probably before things like self-driving cars reach um very high reliability so I would add to this list um I don't know how to phrase it exactly but you know someone who values uh a kind of conversational relationship like a friend with an AI there are people like that already right I mean there are kind of weirdos who really love talking to chat Bots even though they're extremely superficial uh but that's going to change very rapidly I I don't I don't know when this will be but I think it it feels to me a bit like just before internet dating really started like you sort of know something is technically possible but the community consensus is that it's for weirdos and only freaks would do it uh the yeah the public perception around internet dating is is one of the the advantage of being an old person Adam like us is that you see things just go through these stages and it makes you aware of how rapidly and profoundly people's views on things shift it's just like oh yeah when you know when we were young it was just like first of all there was no internet and then when
there was internet it was obviously crazy you're just asking to get you know something have something bad happen to you to date someone over the Internet or to start a relationship geez I mean I can of course we can remember when it was crazy to think anybody would ever buy anything on the internet yeah like I mean use your credit card to pay something for something on a website are you nuts you're just gonna get robbed blind yeah that's more or less like the same category is replying to an email from a Nigerian scammer or something oh yeah I mean in the early 1990s that's exactly how it felt and maybe with good reason too I mean those were primitive days um but yeah I I I think well this is actually a good place if you don't mind if we can find a place on the board maybe um what I'm looking at on the right-hand side to draw the schema that I've proposed a number of years ago um and it's dated now but let's at least put it up there because it's it's old thinking but maybe it'll be interesting and that is you know what what what occupy how to think about occupations that are threatened by Automation and I was thinking purely in terms of automation uh in the past um in other words by narrow artificial intelligence not by General uh whether sentient or non-sentient artificial intelligence I was thinking strictly speaking by algorithmic Nero AI um but at any rate okay so it's it's two axes and you can think of it in four quadrants high low and high low right okay so the two axes it doesn't matter which way around you put them um the two axes to draw uh are maybe I can do it myself well I'll let you do it because you've got the stylus um the two axes are uh uh the the barriers to entry of the occupation okay on the one axis and you can think of that as sort of the educational or skill requirement or or in in broadly and perhaps you could you could because we're talking about AI you can think of the quote unquote intelligence um requirement but that's probably misleading um and I'll explain why in a second using the example of drivers right so we used to think well you know I mean any bozo can drive a cab uh you know can drive a taxi uh no it actually takes a massive amount of uh intelligence and it's a much harder thing to solve than for example um uh you know um well many other many other things that have fallen um uh to automation sooner than driving Vehicles okay so um we have to be careful about how we use that term in intelligence but certainly barriers are
barriers to entry and educational attainment requirements you know so how many degrees you have and so forth okay that's on the one axis and you can think of that you can break it into low and high there and then on the vertical axis as you've drawn it the stakes of the uh the stakes of the occupation in other words and this is this and when I originally wrote about this I used the word Stakes specifically because it's not just about risk it's also about reward right so if the stakes are low um then there may not be all that much of a pressure priority to automate quickly whereas if the stakes are very high in other words if either there's potential for uh catastrophic loss you know death dismemberments permanent injury you know so forth and especially anything that might scale um and on the flip side if there's potentially huge reward or payout then the stakes are very high right okay so now we can take those four quadrants and we can put occupations in there and you can say well it's interesting because there are occupations that fall in all four of those those quadrants um and the place and the reason why I thought this was sort of interesting originally when I published this in maybe 2014 or 15 something like that um was that it's not intuitive right so like you can think of like a you know driving a car the barrier to entry is low but the stakes are actually High and Tesla has for example learned this the hard way right it's to make a safe self-driving car is extremely difficult because driving is freaking dangerous right you're you're a you're a two-ton steel and metal box moving at 80 miles an hour it's an absolute death trap but the barrier to entry to that occupation to driving is very low okay um so the so because the stakes are high the the uh incentive to automate this sooner rather than later is is there also the reward is huge right so whoever succeeds in in automating driving is gonna is going to realize it's ungodly uh return on their investment and it scales right so you crack this once and then there are literally hundreds of millions of uh vehicles in the world to be driven so maybe we should be completely very large the the axes aren't really symmetrics uh we're thinking in the context of automation the barriers to entry is kind of for the human not for the person seeking to automate it right it's and then the stakes you're kind of thinking about well I I suppose it's both in the way you just described it um both the stakes for the for the person who succeeds in
automating it and also for the person currently doing it this seem like rather different Stakes somehow which one yeah but it's vertical axis uh yeah I I guess I guess I'm being a little floppy in both of these in both of these originally the I think if I if I again think back to my earlier self I think my original perspective was I was coming at this from the point of view of of we were trying to imagine which which occupations which jobs were vulnerable to automation first like which were the things that were going to go first which were things are going to get automated which which jobs were the robots coming for first and and who should prepare for to get you know to be technologically unemployed um and so it was very much from the human perspective it really wasn't through the lens of the AI technology itself that I was thinking of this in terms of barriers entry um so that's how I was thinking about this um in this in terms of the stakes yeah I'm so little sloppy you could really have multiple axes here I've sort of I've sort of a sloppily condense them down into a single Dimension but it is with this word Stakes which is a kind of a weasel word um you really should probably break it out in terms of you know um risk and reward and and you know a few different dimensions there that are not necessarily uh perfectly overlapping and correlated but might be somewhat orthogonal um but at any rate that's this was thinking from quite a while ago but I think that the the the the the barriers to entry part was what was originally interesting because people were thinking that it would be the quote-unquote low skill jobs that would be automated most easily and most quickly um and uh what I started looking at it this way it became clear to me that that's not necessarily the case that intuition it was seemed like it was likely to prove false and I think it has I think it has but I don't know if this is again this is old I don't know if this is still useful to us but let me I can put a few things or we can put a few things in these quadrants and then maybe that'll help us you know it really stimulate some conversation yeah maybe we should distinguish between I mean This this term low skill high it's a lot of complexity so I think correct me if I'm wrong but the observation you're making is that you might have an idea of low skill and physical labor as in you don't need a lot of cognitive ability or sort of training in terms of thought processes in order to perform the job
versus you could also say this cognitive labor that's relatively low skilled um and it from the point of view of today's systems it does seem that uh what we think of as low skilled in terms if it's a physical operation I mean people probably nobody who's really familiar with how Trucking Works would would look at it and say it's particularly low skilled I mean you're driving a very large vehicle and you know there's a whole video games about driving trucks and you know they're full of complexity and so on but uh we're just very good at these physical activities because of our Hardware platform uh so they they look low skilled to us but these are in some sense quite difficult activities for the machines who are coming from a very different Universe somehow whereas the cognitive lilos I mean yeah let's not say low skilled but some things we tend to think of as being fairly clever uh are in fact highly automatable um things like apparently what stable diffusion does that's that's what you're getting at right yeah that's exactly right that's exactly right yeah you're going to add some things to the table what are we adding yeah so so let think about things that are low let's just go sort of around the the um around the table there so something that's low stakes and also low barrier to entry from the human perspective from the human perspective let's stick with that because that's how I have these things drawn in here originally so if your barriers to entry as a human employee as a human worker are low and the stakes are low what would those sorts of jobs be that would be things like being a cashier working retail um uh uh uh doing you know janitorial or custodial Services landscaping and and other things that we might call uh menial labor in other words you know labor that doesn't require masterful uh you know thousand hour five thousand hour ten thousand hour um skill building to become um uh uh employed and competent in so this doesn't include things like being you know um a carpenter or a plumber or Builder or or anything like that um these are all really Junior positions that that that um most humans could do okay so that's that's that category um uh you could imagine the maybe just uh to anticipate a criticism I want to say and I know you agree that uh nothing about what we just said or will say is is meant to look down on people in any way uh and it's it's kind of a mistake of our civilization that we judge people in some like with some moral valence uh depending on the work they do and
this is something we should and have to get over uh very quickly um so this is not there's nothing better intrinsically about a doctor or a lawyer versus someone who works in retail they just have a job that requires more training yeah that's exactly right and and the Temptation there is to construct a uh a a highly moralized Merit meritocratic hierarchy around employment and occupations and there are all kinds of of uh philosophical problems with doing that now and and that's that's just leading up until now and then of course going forward many of these philosophical concerns become very much real world practical problems uh when you have the the Specter of artificial general intelligence looming right because compared no human is going to be competent compared to AGI especially super intelligent machines right we're all going to be in into a first approximation equally useless um and the difference between between somebody with a fancy pants job and somebody with a quote-unquote menial um uh job or red night job or anything like any of those derogatory terms that we use those are going to seem hopelessly quaint and and very much much guided in the context of hyper-competent super intelligent machines and uh yeah so we need to we need to it would behoove us to get um to get ahead of that curve on that if you ask me but yes I completely agree with you Dan we aren't and we're not trying to to um denigrate or derogate any of these positions or simply just offering a bit of a schema and again I I um only putting I'm putting those skill requirements and intelligence in that axis in scare quotes the main thing is that there are barriers to entry from the perspective of employers and employers and employees in the real world economy um so uh now you can have let's go to the right because this is interesting you can imagine that there are jobs where the barriers to entry are very high in other words they require much more training much more time perhaps more degrees and and qualifications and so forth um but the stakes both in terms of risk and in terms of reward are not necessarily not necessarily high and you laugh you're writing in exactly the things that I had there I had Scholars and uh uh poets and writers and musicians and artists and all of the things that are in that quadrant where life and limb are not at risk right the life and limb are not at risk and yet there can be very very high barriers to entry into those occupations and professions um so uh that is the sort of thing that
I was thinking that's the story of the sort of things um uh that goes in there um yes exactly uh and if you want to I mean again if you want to say if you want to deride anybody well we know we can look around the chart and see who deserves who deserves a bit of score okay um the the uh let's go um to the top left quadrant sorry Adam you just you stepped a little bit over in the orbcam swung so yeah thanks that's good is that good yep okay so so go to the top left quadrant here now the barrier to entry is low but the stakes are high okay well it's surprisingly there are a lot of there are a lot of jobs that are in this domain and this is where people are sort of woefully underappreciated and exploited so a driving is definitely in there because although the barrier to entry to that that occupation is low the stakes are very high especially risk to life and limb other examples include things like food preparation so people who work in a restaurant right you you can't you can't be Cavalier you have to be hygienic and safe with people's food um and there are a number of of uh positions that are like I think that where the barrier to entries fairly low but nevertheless the individuals involved are taking a serious risk and and responsibility and threat of culpability onto themselves in those occupations and they're often often not nearly compensated well enough so for example teachers are in this in this quadrant because right so and and oh my goodness especially teachers and Care groupers of young children um and also perhaps the elderly where the you know the stakes are enormously High your child's entire future mentioned their life in limb are in a person's hands who is typically grossly under compensated compared to other occupations especially the ones that we're gonna you know that were that are on the right hand side of this uh that have high barriers to entry okay so now the top right these are the top right quadrant are the more stereotypical profession professions right so these are things where there's a high barrier to entry and there are high stakes so these are your surgeons and you know your brain surgeons your rocket scientists people who are building things that can't fail or people die people who are you know providing services and that includes people like lawyers where you know you're you know a good lawyer will keep you out of prison a bad lawyer won't uh or or keep you from going bankrupt or being ruined it for example so an architect has to be
highly skilled to design a building that's safe and and Engineers vehicles that are safe and so forth so these are sort of your stereotypical professions anyway this was the schema that's it uh the the I don't know if this is useful to our conversation or not but um if I dial my thinking back almost a decade this was you know this was what was on my mind and again I think over the last decade what's been very surprising is which things have been have fallen already or begun to fall to Automation and AI uh things I wouldn't have expected like things in the bottom right quadrant um but one could imagine that the the pressure is not equally present on all of these occupations uh you know because the reward or the avoidance of risk from the perspective of an employer is not there um whereas you know you you could imagine enormous value to automating all of the jobs in the top half regardless of the various entry the top half of of the the the space we've got mapped out here anything with very very high stakes if you can automate that you can lower your risk and and on the flip side of that increase your probability of obtaining a scalable very large or scalable reward um if you could automate these things um and uh yeah so I don't know how if that helps our conversation but um we could talk about you know which of these things do we really think are amenable to Automation in light of all the recent developments since I originally thought of this uh schema um which things are really likely to go which things are going to be last to go that would be a another way another question to ask like which which if any of these jobs are are going to be safest for longest yeah maybe that's a good segue of bringing back the the relevance to individuals thread uh that Rowan asked about some weeks ago you know what does all this mean for for me uh in making choices about careers or just what to do um yeah choosing choosing something that's uh above the high tide Mark for as long as possible is probably high on the list of things to think about when choosing careers in the near future I would say yeah I agree yeah it's I mean it's tricky I think here I have to bring back in a topic you've mentioned many times Adam which is um the well the way I would put it is that as at a phase transition uh some quantities sort of go to Infinity so when you have a disruption it may be that some jobs are being automated but a market is growing a new market is growing in size so rapidly that even if
a certain job is being automated if the demand is going to Infinity uh then actually there might be more people needed to do their job temporarily or at least an Allied job May suddenly require many many people so I think I'm quite uncertain about the demand for programmers for example so it's quite interesting to reflect on the the wage premium for programmers in say the United States over the last several decades because it's well it's superficially easy to say well you know of course programmers are highly paid but it's not that obvious right I mean the programming software engineering is a job that's been around now for decades people see the wage premium so they do those degrees and there's certainly a large increase in the number of people doing computer science maybe it flattened out 10 or so years ago but it increased very rapidly but the wage premium remained which wasn't what a lot of people were predicting and that seems to be a consequence of just well software eating the world has Mark Andreessen would say right even though there's a lot of people going into programming as a career the domain of application of that job is growing more rapidly at least judging by the wage premium than the number of people going into it even as that increases very rapidly itself so that complicates these kinds of predictions to the point where it seems very hard to me to really predict things with a fixed like unless things that are sort of scaled to the population say Radiology right the number of people breaking bones or needing x-rays is roughly proportional to the population and their health and the risks they're taking and unless the population goes and starts increasing again rapidly which which may happen but it doesn't seem to be gonna happen I mean that's a very slow process by definition almost um so the demand for radiology probably isn't going to change unless there's some new reason to take an x-ray uh but that does seem Maybe not immediately but over time will be pretty subject automation at least large parts of it so there's things like that where you can fairly confidently predict that is not going to be a smart career but some of these things like programming or um teaching is another one right so you could certainly assume my son goes to synthesis school which is some online kind of class and they just emailed us this morning about a new AI tutor their trialing okay so uh today in the code seminar I'm going to be implementing an AI bot that'll stand in meta Union you can talk
to it and that kind of technology is going to be uh widely used for tutors and teachers and we will scale the automation of teaching and tutoring like crazy over the next five years but still maybe there's even more demand for teachers because we shift our civilization to actually giving a about teaching in a way that we don't currently so that it just may be that we choose to spend a lot of human resources on it even more than we do now so maybe teaching actually is a more desirable and well-paid profession than it is today so a really fun find it difficult to give advice on you know careers I don't know what your take is on that uh foreign yeah I mean I think my take is very similar to yours the one thing that I would add there is that um there are there seemed to me to be occupations that have relationships the full complexity and dynamics of um Human Social relationships embedded in them in other words they are personal rather than impersonal uh uh services and a teacher is a great example of that other sort of sort of ones that are commonly mentioned are things like coach um uh therapist and and um and so on um where it may be more difficult again not impossible necessarily I mean you were mentioning earlier the relationships that people have with chat Bots um but it it may be the case that that is a that fulfilling the the entire uh you know bad the the full bandwidth channel of a a human to human personal relationship is is is um takes longer and is more difficult for AI especially that isn't sentient to to um to supplant and so I think I can imagine a scenario where we very uh dramatically humanize a lot of occupations that have been under pressure to dehumanize for a long time and uh and these are occupations in some instances that may be very old you know they may they may extend deep into history and that that it were dehumanized relatively recently and it might be it's coming it might be fun to imagine that they might be rehumanized um and The Optimist in me and the sentimentalist in me uh thinks that a lot of things might be like that you know we might go back to spending more care more time and more attention on each other and on Small Things um uh that we once maybe spent took more time and care with that that uh you know in in the past um so yeah and and so teaching is a good example um uh but even you know so taking it take something like like a landscaping I mean there are there are people uh who really enjoy just the the the time and the Solitude
maybe in some instances something it's just a social dimension of um being doing be being very meticulous about caring for things gardening is an example right um and yes you can you could automate gardening fully and just and get the results but the the the final results is not identical to the entire process and the experience of um doing the work of that occupation and then enjoying the final result at the end of it and um it could be that that more work will be treated with that sort of depth and care and perhaps built into one's identity and experience as opposed to just being you know um alienated and turned into toil and drudgery and I think there are cultural examples around the world where this is the case that you know uh in my own family because I'm married to a person of Japanese ancestry um I'm a little bit familiar with Japanese culture and treated more traditional Japanese culture and still somewhat today um does provide constructive examples for how to how to take a great deal of care in seemingly small and mundane things like preparing a cup of tea or um or working in the garden and um it might be that there is there could be a general shift in our uh in Humanity's posture towards work then um uh as a result of the rapid and widespread automation of lots and lots of um occupations so we'll see um so there's that to think about as well those those two things our personal investment in the work and then also our interpersonal relationships if if that content which has been driven out of a lot of uh occupations maybe it could return to them or or expand within them within those occupations that still have that that Dimension to them hmm there's a few comments in the chat I'll just read out to inform the discussion yes I'm Matt points out that people have been talking about um disruption in education for a long time it may be different this time maybe not and Rowan made a good point which is that uh less people with less jobs and well less people with jobs rather might reduce the need for professional teachers since parents will play that role uh yeah I can say that's personally happening in my family so we are basically choosing to homeschool Russell um and it's only possible because of first the internet and online tutors and uh more and more of a role of AI tools as well I expect over time otherwise we simply wouldn't have the time to do it and it it does mean that we spend more time with him than we we would otherwise so I think that whether it's
remote work or AI tools the coming disruptions will allow people to resolve coordination problems that were only possible to solve in a certain way before so I think we'll see a lot of that where it's it's not even even slight changes in what's possible can cause new solutions to coordination problems across many people and many different aspects of Life open up new solutions that are really quite different uh so that's of course where new occupations come from I suppose right after disruptions it's just that the the landscape of things that people want done will change as people live differently so uh there's a much larger market in the future for online tutors than existed in the past simply because a lot of people will start doing what we're doing already are [Music] um so this is one way in which that's not only AI but uh things will change and Matt is um commenting that yeah maybe we should well I'll give it a spin maybe we shouldn't be too Keen to just go do gardening and let AIS Run the World um maybe maybe that's not the optimal outcome for human beings depends depends how serious the gardening is maybe I tend to think of my role at meta unism this is more like a gardener than anything I don't know that it's not one of one of the things that that is always a danger in a conversation like this um is the what I in in the paper Dan I'm sure you remember it wrote it a number of years ago I talked about some of the what I call informal fallacies so sort of consistent laws in in problems through current errors I think and thinking one of them was the arrival what I call the arrival fallacy and that's the idea that you know that we're going to arrive at some point in the future and that's what the future is going to be like you know and and even if we tell ourselves well you know that won't be like that forever the thinking is still well we're going to arrive and then there'll be some stasis there'll be some something static you know we'll be in some homeostatic conditions something relatively stable for a while I think that that's a again that's that's kind of a dangerous assumption to make it may be within some domains there's some truth to that but for example um in in all of this you know we're still thinking well you know AGI is going to arrive and then how are the jobs going to change and then you know what are we going to do and and so that may be perfectly fair and if Sam Altman is right and the the there isn't you know sort of a a hard takeoff of AGI
um and it's a little bit slower than we imagined especially if it's not sentient then um okay fair enough you know people are still going to have to have a job they're still gonna have to work you know it isn't going to be overnight that societies are going to be wealthy enough to start paying out you know ten thousand dollar a month ubis um and so that all of the humans in this Society can immediately retire and go about their Leisure and so forth we know that that's not going to be the case um but we I think we nevertheless we should probably be pretty cautious because the the time between you know sort of uh it's some sort of takeoff in AGI arriving and then something closer to you know getting past the foothills of the singularity to the actual Singularity um is probably not gonna be that long I mean I'm thinking 10 years at the at 15 at the outside and um you know I I that's even in my life that's more than twice as that's more than twice as long as my adult working life now basically I've been working for more than 30 years um working since I was 15. and so I mean that's not very long uh if that's really the case and then if everything becomes moot after that you know whether nobody's working or jobs don't even mean anything and you know people are mostly in the metaverse if that's a thing or people are you know merging cybernetically with uh it's super interesting with an artificial super intelligences all that Singularity stuff right so anyway I just want to dial it back to this this this um uh the to to reframe it whereas what am I trying to say here what I'm trying to say is all of this conversation needs to have a date to it we need to recognize that whatever happens in with the things that we're talking about here this is this is only even applies to maybe a 20-year 25-year period and then you know it's it's out the window anyway um and with that in mind then you know that that Justin itself ought to shape the conversation a bit right hmm because now what not like we're talking about Landing here in some new condition and then okay we're going to be there for three or four generations and we need to really figure out how to you know organize accordingly around it that is not on the cards yeah I think that's right and implicitly uh if we're talking about safe jobs the way I think about it is there'll be an interim period I mean okay if things go very badly we're all just did uh if there's a kind of quasi-utopia on the other side or at least some very rich
State and ubis and you know things are kind of okay all very good uh still the interim period you know you don't want to lose your job in the next five years nobody's going to come and help you right you'll just not have a job and if you're American you'll just be left to die in the streets yeah right sorry I'll apologize too many of us feel that way so okay then it makes a difference whether you get on the right boat now right so there will be an interim period where things are going to be very Rocky uh things are going to change very rapidly there'll be a lot of negative responses people pushing this way and that way it's going to be chaotic and difficult and maybe dangerous and then we come out the other side and things will be you know it won't last forever but you want to be in a good spot for this period And I suppose my own career choices and movements and decisions over the last five years have more or less been largely around that uh in the name of you know uh not wanting to to get on put on the rubbish Heap at some point um let's we've got about 10 minutes 15 minutes and we should um move on to the uh tea break I want to show you there's a new building uh here in Medi uni and I want to I want to show you what that looks like um yeah maybe let's talk a bit more about this safe jobs idea uh I was thinking I'm reading a new book by Ross Garner about superpower the sort of idea of transforming the Australian economy along the lines of the opportunity in renewable energy and yes here's an idea I want to see what you think about it um it's not like the robot manufacturing computernium world will arrive instantly right even if that's the direction we're going with you know very capable widespread robotics etc etc it will need a lot of energy and maybe eventually the robots build and install all the solar panels and you know design and build and install all the power cables and all that but that that isn't coming in five years I don't think so as as this transformation progresses the demand for energy and raw inputs into this growing technological Marvel slash horror uh will be intense so anything feeding into those inputs will be pretty safe and I would think highly valued for some time so for example there's a lot of jobs that are going to exist in Australia building out renewable energy not only the hardware and the engineering but also the coordination and management and designing all those markets and making sure they run well and there's okay there's there's books including the one
I just cited which will tell you those kinds of careers which are you know several of my students and others I know have gone into those kinds of jobs those seem to me like relatively safe because the demand for energy whatever is going to happen I mean even in a scenario where there's no AGI it's going to go up like crazy and if we are running all these AIS in data centers and they have ideas for all sorts of amazing new things to do the demand for energy will be even higher so I would say that that kind of infrastructure and everything surrounding it is is kind of going to be a bottleneck and therefore something that's highly valued I would say the same about coding actually um machines are very rapidly making progress on coding I think there's an estimate I don't know which organization it was but so Google cited its internal version of the GitHub codex model as increasing programmer productivity across the organization by six percent so this is a using models that write code for you basically and six percent's a lot uh productivity growth of six percent in a year at a company the scale of Google there's nothing to sneeze at and more is coming so a lot of programming is getting automated uh already um and I would just add to that one presumes that the the quality of software engineers at Google is very high relative to the industry and global average and my my naive intuition would be that that the better the software engineer the closer they already are to some optimal amount of performance um so the yeah the idea that you could get a six percent premium on top of a Google engineer is pretty amazing um if it were you know the engineers who write the code for certain services in my city for example well I can imagine giving them a big bigger boost than that with even the tiny little bit of AI help poor guys are struggling with it but in a place like Google that's astonishing that's astonishing and one I mean one okay these models is they currently exist make many mistakes they have many issues for example they just copy and paste in code it code out of Open Source libraries that shouldn't be copied and pasted and I said there's a whole bunch of issues around them and they're quite flawed in their current incarnations and so on but there is no reason to expect they won't become much better and I think it's more a comment on actually most programming work being much more mundane repetitive and pretty low cognitive labor than you might think so I mean the Google Engineers I know
has been probably half their time just plugging things into each other with Proto buffers they're just chugging data around from this thing to that thing and doing some pretty you know pretty boring repetitive Transformations and it's not that much of a surprise that you can automate a lot of that uh so okay but given that it's going to be automated why does it nonetheless seem to me fairly safe in the short term well the leverage and power of somebody who can Marsh all the resources that the AIS will provide which will start off in the digital world in their world of code the opportunities that will be unlocked by that will make the opportunities in software deployment over the last few decades uh look pretty small in comparison I would say right so at least um that's one of these phenomena where the automation is happening at the same time as the scope of the activity is expanding by an order of magnitude so I think that for some time uh still these will be pretty safe and high value jobs can you think of others Adam or anybody else well like just a quick a quick Counterpoint to that I just will throw back to a couple of weeks ago we were we were talking in our conversation about how I think it would I think the example we used then was chess how humans plus AI were were for a brief brief moment uh more capable than AI alone but that didn't last very long as my understanding um and so I I I I do Wonder if you know we don't want to fall into the Trap of assuming that same sort of thing that seemed to be the and I think you had some funny ways that you described that it was sort of the look at this this the copium of our age or something like that I think it's how you described it so we want to make sure that we're not that we don't aren't making that same mistake with software engineering and tools like codex um uh because maybe the debt is just a very unstable equilibrium it doesn't it does that condition doesn't last very long um but uh uh but I can't speak to software engineering because I have so little experience in direct contact with that um but I tell you what um maybe here's a here's a useful how about this it's a useful shorthand for making predictions about uh other occupations that might be amenable okay um think of occupations today or or tasks even never mind occupations which is you know an occupation is a whole cluster of organized tasks together but you could think of it as you could break it down into the into individual or tasks to subtest if you want but anyway
let's stick with occupations okay imagine a line of work or an occupation that you do today that for which it's it it would be or it is really helpful to have a second human being there giving you a hand right so there there are many things that are like that where you know we're we're two heads are better than one we often say we have Expressions we have idioms for that for for capturing that that sort of thing you know or it's nice to have an extra pair of hands you know that that kind of thing we have these Expressions well um perhaps there will be a time who knows for how long exactly but in the in some interim period where uh AI and then one would imagine AI instantiated in some sort of general purpose robotic uh uh platform you know whether it's in uh anthropomorphic uh robot or not who knows but one could imagine um machines basically being able to step into that role of helper or assistant or just partner or peer um to do work or or achieve tasks or or pursue goals um that we know that we do currently do not well maybe in formal occupations maybe even just Hobbies whatever it is um whereas really really helpful to have another person there giving you a hammer maybe even two people they're giving you a hand helping you out and if that were the case we could see we could think through that lens and ask that question well right now it's it's really great if you're a you know um if you're a master plumber to have an assistant with you on site because sometimes you really just need four or five hands and not just two to get you know to get just to just even if the robot is literally like just standing there holding a wrench within your reach while you're you know on your back upside down under a sink or something like that that could be really helpful um so I can imagine a number of uh uh situations that are something like that um and you know this is a little ways out this is this is not in the next five years but I you know in the next 15 years I can imagine um people putting uh machine labor to use in that sort of um uh I want to say constructive but sort of sort of um uh uh uh uh additive as opposed to in an additive way as opposed to in a a replacement way so you get sort of a a non-zero outcome as opposed to sort of a zero come exchanging human for machine labor outcome if that makes sense yeah and I think that does yeah I think your comment and Rowan's making a a comment that I'll read in a similar way uh regarding coding I think he's saying it might be helpful to distinguish the
task of writing code from the broader design and Engineering Process and I think that's a good point so uh especially at the scale of something like Google the task of implementing a new product or feature is I don't know what percentage but if you think about the number of hours that go into it and the team and what they're doing in any given moment it's maybe half code or less read a lot of what they're doing is coordinating across the organization figuring out what to do what is necessary and it's um yeah a lot of process goes around you know also checking that it's doing the right thing and putting in place automated systems that will verify it and all of that is not necessarily coding and that part I mean figuring out what to do it'll be you know we'll be automating low level coding and medium level coding sometime before the AI can run a company and just figure out what product to make right so I do think that and you can see it already with um the new capabilities of language models have led to a a large number of startups that are trying to exploit those abilities to do new and interesting things and that AI tutor I just mentioned from synthesis would be one of those so that's that's a place where humans will find a lot of value for at The Cutting Edge for some time right that's taking these new capabilities and kind of running with them at the level of design and coordination and and all of that um let me let me describe a sort of again a sort of a Counterpoint or Flip Flip that invert that a bit um I don't know maybe this is silly or it's too narrow but let me let me give it a shot anyway so I I I can imagine a situation where the where today I would pay an expert like an architect or a design I'm going to give you a narrow example of Home Improvement okay so today if I wanted to do some Home Improvement some projects to a home improvement on my home um remodel the bathroom expand the kitchen you know fix something in the backyard whatever um uh they're they're there's a design in engineering sort of Dimension to that and then there's the implementation of the design by what in the United States we call contractors and those would be carpenters and Builders and and people with with different sets of specific expertise if you're remodeling a bathroom or a kitchen you would need plumbing and electrical electrician and that kind of thing now if you're handy uh or reasonably handy or or just foolish I'm foolish I won't say handy but I'm foolish um and and masochistic and you enjoy
doing that kind of work yourself um and you can save a boatload of money too because that that expert labor plumbers electricians and so forth are very expensive to the to the end user certainly um then uh the only not the only but the service that I hire in is the design that's the service that I hire in and then I implement the you know the grunt work myself um for better or worse and I could imagine getting the design work from an AI right uh and then and then still doing the you know the quote unquote hard part uh myself um so I don't know if that is exactly the same or if that's somehow the opposite of what you were talking about with with you know design event of of software being safer or or less amenable to to AI but anyway I'm just giving a kind of a maybe a parallel but somehow inverse inverted uh or Bizarro World example alternative to what you're talking about um yeah I would I would agree with that except if you were trying to design a kind of building that had never been built before or uh an exotic uh Music Hall with some amazing sound properties you're probably not going to get that I mean the plan for your new bathroom you can probably get an AI to make that or even your house if it's swallowing fairly standard lines but uh yeah I get that's an implicit point in what I was saying about design and coordination if you're making a new product or a new company if it's a problem that really doesn't have a lot of precedent you you for some time you won't be able to get the AI to just generate a reasonable solution to that problem yeah that makes sense okay we're going to move over to the new building in a moment um wait yes we're waiting hey I can't invent stuff that's not what I said for some time it'll be much cheaper and easier to get the AI to write code that's mostly been written before come up with ideas that's interpolating between things that already exists it will be somewhat later where the AI is saying here's a business idea the people you should hire and the architectures you should use and then directs the staff to implement it or the AIS to implement it that's a timing claim yeah that's right I absolutely think that's coming all right uh we want to wrap up this discussion for this week are there any final comments I mean we can continue to discuss but maybe this is the end of the sort of more formal part of the discussion and we'll move over and have a look at this building and and we can keep chatting and take questions are there any final comments you want to
make Adam before we do that no no more comments for me this is very interesting thanks cool all right um so if you want to detach from the orb uh and follow me everyone so if you can see it already from up here if your draw distance is sufficiently high you can see it in the distance on the other side of the knot over there so that's where we're headed there's a building [Music] foreign [Music] yeah so this is a piece of architecture that uh uh we bought off a um a robloxian coral rhyme so I didn't build this this is much beyond my my abilities in Roblox studio for now and probably forever I don't know how long this took hopefully we'll have a chance to kind of ask some questions of Carnival rhyme at some point yeah it's it's definitely more demanding over here hopefully I mean I hope it didn't feel too laggy when you were back over on the peak did it as the performance of TRS feel any different to usual no yeah okay so that's that's because of game streaming so that's um this is made up of many parts this building but when you're over on the peak you shouldn't feel any influence of them um so yeah I don't know to what extent I mean performance is important so I don't want to put boards in here and then have them be really slow if you're using them for example but I can imagine using these environments for for some things so I'll show you the lecture hall if you follow me [Music] so in the past we've kind of not used indoor environments as I've said before partly because I just really hate seeing people talk about the metaverse or Virtual Worlds and then stick people in the like some very one-to-one reproduction of the terrible building that they they have in the real world as though there's no opportunity to do something different in the virtual world uh but having said that uh you know sometimes it is cool to be indoors and gives a different vibe to things so um we're interested in um and exploring that for example it's kind of cool to look out the window at the world I think yeah I don't know that the audio is any different in here I haven't noticed that um I think the on the roadmap for audio in Roblox is they want audio to behave differently in indoor spaces and to bounce off surfaces in a different way and all that but so yeah talking about future jobs virtual architect seems like a cool job to me building virtual spaces that make people more productive yeah holy moly and if you know the sky is the limit like your imagination I mean even not even your own
imagination is the limit because of course you're gonna hit your imagination is going to be assisted by by uh the AIS that are powering you know the Dali 2s and stable diffusions and mid-journies of of tomorrow and and so forth um so yeah I mean it's kind of astonishing to imagine what will be possible um to produce yeah it's uh I think I mentioned it a couple of weeks ago I am gonna have to go back and and see I'll I have to dig up whatever that Star Trek the Next Generation episode was where the children were sort of effortlessly producing stuff just by imagining it with these with these tools that were cheating they were shortcut I have to see I don't remember what the other what lecture the Enterprise crew gave to them about to these people about how it was how it was the wrong thing to do um I don't know how well that logic is going to hold up now that it's basically a reality or we're very very close to it but at any rate the the yeah it's the the potential for creating for just unlocking unleashing creativity and then explosively growing the space of possible experiences that people can have is amazing I mean that's the promise the the beauty and and the utopian magnificent vision of the metaverse right is this there's no limits it's just your imagination there have been people have been talking about this for a long time you know William Gibson and uh and Philip K dick going back a long way and then you know of course the Matrix exploded the the possibility into the public imagination um and more recently you know we've got other uh media and popular entertainment versions of of virtual worlds and you know we're now looking at seriously creating them but it really is about you know it's it's about what you can create and and who you can be and what experiences you can have in these environments right um uh and that's it's it's it's it's almost I think that was I think what I I sorry to go off in a little bit of a tangent here but when I first was following the news about nfts and the metaverse well I found that very discouraging uh and and it's struck me as very dystopian and I think it may be a lot of people felt a sort of similar way because instead of opening a new door onto uh experiences and um uh things that you could be it was just more of what you could own and what you could have and possess and and that that that you know that was just so stomach turning to me um very empathetical to what I've always dreamed the the metaverse or the the virtual reality could be all about
um and I hope that that's not a silly old-fashioned uh way of you know idealizing what this technology will be used for and I'm sure it'll be some mix of the two of those things but at any rate being in a place like this seeing this skill come to life as it were um is this thrilling it's absolutely amazing so really cool yeah it's um one of those things where in principle it would I would much prefer if this uh if there was like an open source version of Roblox uh that was uh you know the the monetization of Roblox is not really very intrusive for us uh because it's not like you have to you know you don't pay to join you don't pay to sign up for Roblox um so it isn't really influencing what we're doing so far but it's just unfortunately you know people haven't been very successful in building uh voice uh apps online that are at a quality that uh is compatible with zoom just because the problems of designing software that does reliable low latency high quality voice chat is is just really difficult and not something you can even really do as a small organization um yeah as Matt's pointing out to use voice chat you are giving up some personal information which um does come with costs particularly or the possibility of it um being misused or leaked or stolen or or whatever may happen um but yeah like the the infrastructure investment that Roblox is making for example just to make this voice chat work at scale uh and to make you know the places stream to you at low latency and and all of that it's it's really hard for open source to compete with uh yeah that's right um or or what is maybe pretty similar if you're a Chinese citizen and you're overseas you can't show Chinese ID to Roblox they'll just say get lost to a Chinese you can't use Roblox uh which includes some of my own students right so we've had to find workarounds for that so this is it's far from optimal but um yeah I mean as long as the benefits outweigh the costs for enough people will will keep doing it and hopefully there are less intrusive ways of signing up for voice chat in the near term that would be excellent work on it oh yeah that's a good idea I think I mean that's one of the ways in which open source needs a distributed way of collecting coordinating and deploying Capital otherwise it's not going to be able to compete with the whether it's AI or metaverse or these new trends that are coming we're going to see a new wave of ways of deploying immense amounts of capital right and the way in which
open source or Academia can't compete with companies for work on the metaverse or AI is going to come to many many other areas of human activity so it's yeah but I mean I'm talking about building data centers there's no like building a metaverse is not something that is just writing code you need immense amounts of hardware and I mean Roblox is building their own data centers all around the world to make sure that there's low latency connections to as many users as possible and those data centers communicate directly to each other in very high speed and that kind of thing and that's I mean it's it's maybe not such a big problem to have five people in a world but if you want to have 50 people in a world using voice chat and seeing the same things and interacting in a way that feels natural this is a highly difficult distributed systems problem which okay maybe it's possible to solve it in an open source way but it's not a it's not like writing a Linux kernel this is a problem that I don't think has been solved in a comparable way outside of centralized control in large companies yeah but the internet uh that's a I mean what do you mean by the internet you mean the protocols or the actual Hardware the the landlines are owned by telecommunications companies largely rent if there was a country in which the underlying fiber was somehow coordinated and deployed in a community fashion uh then that would be analogous to the kind of investment that is required to have a truly open metaphors I mean I guess that's the direction which some of these crypto related projects are trying to go but they sort of seem to become very easily captured by kind of scammy plant the flag look to profit Behavior so it's I have high hopes for cryptocurrency as a new way of coordinating large systems of people doing good things but so far I don't see a lot of evidence of that being done in good faith unfortunately uh yeah any any points people want to bring up about the topics of today or questions and yeah that's right um I was wondering about it's something we discussed maybe two or maybe three weeks ago um about sort of looking for signals from different groups about their point of view on when AI is going to be a thing right um and I guess I was sort of thinking I guess I haven't fully fleshed out these ideas but even if you um you know so you think there's maybe a five percent chance AI will happen before 2030 or something like that it still might make a lot of sense to have enormous Investments like very
large investments in it because the cost of missing out if you're wrong or if you think the most likely scenario does play out is much much larger than the cost of losing your investment if what you see is the most likely scenario does play out yeah you sort of yeah I think that's true I yeah yeah and um in a funny way uh I don't I haven't thought of Flesh this thought out but I think people might be misreading what's happening with facebook slash meta so it like Mark Zuckerberg has talked about they changed the name of the company and now it's about the metaverse and so on but if you look at what they're actually doing what they're doing is spending enormous amounts of money to buy gpus they have one of the largest supercomputers on Earth thousands and thousands of the latest Nvidia gpus so whatever they're doing with the metaverse it involves I mean the metaverse as they see it will have a large AI component right you'll walk into a space and it will sort of build itself you know in many ways be it AI heavy so you could say they're a metaverse company but really it's you know the the asset they have is a lot of gpus so you could read the what's going on with Facebook as being a large bet on AI and of course the large tech companies are doing that um Tesla is kind of another big stock that's a bit on AI not doing so well lately I suppose um but yeah I would right I would expect as you're saying that as people's level of certainty about this Rises you'll see fairly crazy I mean it's already pretty crazy level of investment in AI companies yeah I don't know if you're reading it but the um Ian Hogarth and um and others I forget their names every year put out a state of AI report which talks about the level of investment in AI startups and you know what's been happening and and the Russia funding in I mean it's it's hard to keep track of actually because it's happening in so many areas of science but level of investment in um Ai and uh in Pharmaceuticals drug Discovery chemistry Material Science is even in just over the last year has really skyrocketed and also in biology so in biotech those those companies are there's a bit of a reckoning in biotech recently but it's just in the last six months I think two or three significantly funded startups looking to apply large language models AKA Transformers to to um drug design and a protein design synthetic biology this kind of thing so yeah those those bets are being made for sure I guess what I'm asking is like do some
of those bits still make sense even if the people making them don't really believe that AI or a AGI maybe is going to happen soon just because yeah I think so there's a chance they're wrong you know what I mean like yeah it's probably how VCS think about it I'm not sure they're sitting there thinking it's definitely going to be here by 2030. yeah that's right as Matt says uh the uh the Dutch book justification of Bayesian probability is you believe whatever you're willing to bet so I guess uh that's that's appropriate but it's weighted by like the potential loss of payoff right yeah well the the other the other thing to think about um and again this is sort of like uh I mean this is what you don't encounter in any betting real world betting environment not real world sorry in any in any artificial betting environment like in a casino because the house is odds Stacks odds against you but you can imagine in the real world maybe there's a roulette wheel that has 10 10 slots on it and you know only one of them can win but it's going to pay off a million to one right of course you put your money on all 10 because you can't lose right so it it if if you if if you think there are a small number 10 for example or maybe less leading candidates who are likely to you know crack the proverbial code on AGI well you simply bet on all of them and it doesn't matter who wins the fact that somebody is going to win is going to multiply your investment by more than the then the denominator you had to divide it across to invest in every single course in the race does that make sense so unlike embedding in a casino it can make sense to put money on every reasonably plausible candidate um and and you can't lose so that to me I I think that I think it's very likely that will happen if it isn't already certainly happening or written if it isn't certainly already happening already um the outliers there the wild cards there that um you can't invest in every one of these for example you can't you can't invest in clandestine government projects that are aiming at developing AGI and so if there are if such projects exist so um but if you could then it would be very rational I think to just spread your investment across an entire industry and yeah you know if you're not if you don't let it spread too thinly because there aren't that many horses that are in the race well then it's really hard to lose so yeah on that note I think we'll wrap it up for the day thanks everyone and um okay yeah we'll continue next week and