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.


Adam discusses the potential of AI and automation to revolutionize the way we work and create a brighter future. He references economist Tyler Cowan to discuss the use of AI tools to give knowledge workers a competitive edge. He also outlines a three-step formula for success, making a joke and referencing a documentary about a famous artist in the process. Adam suggests that AI can uncover novel ideas and insights, but still can't replicate the unique properties of humans. He advocates for the adoption of AI tools to keep knowledge alive and for the potential of technology to create an interactive and intelligent future.

Short Summary

Adam is discussing the amazing results of his work on GPT-3, an attention mechanism used to process information. This tool has the potential to save time and effort for organizations by rapidly digesting and understanding documents. AI has the potential to automate tasks and replace human roles, but it can also do more than that. AI can be orders of magnitude faster than a human, transforming productivity and prosperity. This disruption has positive implications beyond just the replacement of human roles, opening up the potential for a more prosperous future.
The speaker is impressed by the capabilities of AI and automation, arguing that they can revolutionize the way we work and create an optimistic vision for environmental disruption without sacrificing human prosperity. They mention Tyler Cowan, an economist at George Mason University, who is an early adopter of technology, and discuss the use of chat GPTs for idea generation. The speaker believes that these tools can give knowledge workers a competitive edge and notes that it is difficult to generate insights comparable to those of AI. They advocate for the adoption of these tools, as those who do not will become uncompetitive.
The author discussed their progress in AI and automation, then explored a three-step formula for success. They used a joke to make a reference to a book about not hiding in a cave during a storm. They then discussed a documentary about a famous artist who created artwork for a version of Dune that never got made. Chat GPT and other generative AI tools are able to reproduce content, but have not been able to generate anything fundamentally novel or new. The speaker expresses scepticism that something fundamentally new can be produced by simple recombination of data, and suggests that breakthroughs have happened in the past. They suggest that drawing insights from noise or mathematical structures may be more likely, and that the idea of something special and magical about human creativity is likely to be humbled by the realisation that most art is simply stolen and recombined.
The speaker suggests that AI could be used to uncover novel ideas and insights, but humans still have unique properties that AI cannot replicate. GPT3 utilizes existing knowledge and allows conversations to be more productive and humorous. It is also suggested that an anecdote bot be developed to animate knowledge and keep it alive, and a real-time transcription and captioning bot be developed for more convenient conversations. Finally, the speaker references an episode of Star Trek The Next Generation for viewers to look up.
This transcript discusses the potential for technology to help expand cognitive capacities, such as neural implants and high bandwidth interfaces. It is suggested that stories help people understand and contextualise information, and AI agents can have an inner life that persists over time. The speaker is inspired by the current level of technology and its potential for the future, drawing a comparison between an interactive book in a book and what technology could be like in the future. They express awe at the potential of the future and the idea that when everything around us is intelligent and can tell stories, it creates a new experience.

Long Summary

GPT 3.5 is a language model that is prompted with text every 10 seconds. It takes its observations, past actions, and what it has said and generates its next action, which is then transformed into an embedding vector. This vector is the thought vector of GPT 3.5 and is used to look for similar vectors in an online vector database called Pinecone. This vector space is 1536 dimensional and is used to find pages of a book that are related to a particular topic. The pages found are then added to the prompt and GPT 3.5 can use them to form its thoughts and respond to questions.
Adam is discussing the astonishing results of his work on GPT-3, an outer layer of the internal attention mechanism that GPT uses to process information. He is motivated to continue working for RethinkX by his optimism and faith. He encourages the listener to not be cynical and to live by optimism. They then discuss what they should talk about today, and Adam expresses his admiration for the cool results of his work.
The speaker is impressed by the ability of a model to digest and learn from a corpus of material. This could be immensely valuable for organizations, as it would save a lot of time and effort in reading, digesting, repackaging, and drawing insights from documents. The speaker hopes that the model can be used to replicate the work done for their team and then expanded to other material. This tool has the potential to be incredibly useful for millions of people and organizations.
AI has the potential to automate tasks and replace human roles, but it can also do more than that. AI has the potential to do tasks orders of magnitude faster than a human, such as reading and digesting a report. This would be more disruptive than just replacing a role, as it would do more than that. For example, a bulldozer can move a huge amount of earth compared to a person, and AI can do tasks 10,000 times faster than a person. This capability would be transformative in terms of time saved, and would be a more disruptive feature than just replacing a role.
The potential of automation to replace human labor is a topic of discussion, with discussions on how it could free humans from toil and drudgery and open the door to positive futures. The idea of AI and automation being able to do things much more quickly and potentially more accurately than humans has implications for prosperity, beyond just the replacement of human roles. This idea is consistent with other disruptions, where the increased speed and accuracy of machines has a different, more positive impact on productivity and prosperity than the mere replacement of humans.
RethinkX has a standard line that disruptive technologies are not just substitutions for older technologies, but provide a greater value and capability. AI and automation are examples of this. They are fast and provide advantages that people underestimate. This technology will push on our arrangement of labor and coordination differently to other areas. As an example, a one and a half hour talk was distilled into a five word title and a paragraph by chat GPT. This shows the potential of AI and automation to revolutionize the way we work and create an optimistic vision for environmental disruption.
The speaker argues that modern environmentalism is characterized by pessimism, believing that sacrifice and downgrading of human prosperity is the only way to achieve sustainability. However, they believe that environmental problems can be solved with innovation, new knowledge and ideas, and science and technology. This assumption undermines confidence in the idea of progress itself, but the speaker believes that these solutions can be achieved without incurring huge losses of human prosperity and quality of life.
The speaker is impressed by a chatbot's ability to quickly generate a script for a radio interview. They discuss using the chatbot to generate titles for YouTube videos, and suggest a clickbaity title for an environmentalism talk. They mention Tyler Cowan, an economist at George Mason University, who is an early adopter of technology in his 60s. The speaker is amazed by the speed and ease of the chatbot's capabilities.
Cowan has observed that conversations are effective in refining thinking, and that chat GPTs are becoming increasingly useful in idea generation. Conversations are scarce, so chat GPTs provide an alternative that is fast, cheap, and always available. This changes the equilibrium for interacting with other humans in the idea generation process. Sam Harris emphasizes the primacy of conversation, which has been a part of democracy and science since the Enlightenment. Now, individuals have access to fast conversations with chat GPTs, which allows them to refine their ideas quickly and effectively.
The speaker is discussing the potential of AI tools to give knowledge workers a competitive edge. They note that it is not perfect, but it is cheap and fast, and they feel that those who don't adopt it will become uncompetitive. They have already begun to lay the groundwork for acquiring the capability, and they will need to grapple with the implications of these new tools soon. They also note that it is difficult to produce genuinely novel and interesting insights that are comparable to the ones generated by AI.
The author discussed how they have not achieved enough in the AI and automation space to be able to publish anything yet, but they are hopeful that their tools will help them make more progress. They then discussed a three-step formula for success. A joke was made about a cave, which reminded the author of a reference in a book about not hiding in a cave during a storm. They then talked about a documentary about a famous artist who created artwork for a version of Dune that never got made, and how the aesthetic of other films, such as Alien, owe a lot to the artwork. The author is hopeful that their tools will help them make progress in the AI and automation space.
The article discussed the story of a version of Dune that was never made, and the moving experience the creator had with the body of pictures. GPT was used to generate an insightful comment on the juxtaposition of Yoderowski and AI in filmmaking. Yoderowski was interested in the alchemical film, in which the process of filmmaking is seen as a transformative journey for both the creators and the audience. AI technologies present a challenge to this motivating interest, and GPT's comment asked if the resulting film is still alchemical in any sense. This comment delves into the heart of why Yoderowski was interested in film and what the implications of AI technology in filmmaking are.
Chat GPT and other generative AI tools are able to reproduce content, but they have not been able to generate anything fundamentally novel or new. Humans are innately attracted to novelty, which is why it is important to use AI to search for new insights and ideas. AI can help us to recognize value and content, and to find new ideas that may have been missed due to lack of time and attention. This is a form of due diligence in the space of ideas, and is an evolution of novelty.
The speaker expresses scepticism that something fundamentally new can be produced by simply recombining data in training sets. They suggest that the idea of something special and magical about human creativity is likely to be humbled by the realisation that most art is simply stolen and recombined in simple ways. However, they also suggest that breakthroughs have happened in the past, such as Jackson Pollock's work a century ago, and that this may have been something that could not have been produced by simple processes. The speaker is sceptical that human creativity is something relatively simple, and suggests that drawing insights from noise or mathematical structures may be more likely.
The speaker suggests that there may be an explanation for the aesthetic appeal of Pollock's art, based on its fractal dimensionality. He is skeptical of this idea, however, and believes that AI will soon be vastly more capable than humans at creating novel insights and discovering new things. He does not believe that humans have any special, magical property that will allow them to remain the source of these insights and novelties, and views it as both desperate and arrogant to think so.
AI is becoming increasingly advanced and is able to do many things humans can do. However, humans have a different sense of humor and art that AI may not be able to understand as deeply as humans. We are in a "centaur era" where humans and AI can interact and create insights faster than ever before. AI is becoming increasingly popular in many areas, such as here at Median. AI is able to provide more knowledge and insights than humans, but it is still limited in its ability to understand the deeper aspects of human culture.
GPT3 is an AI that is capable of drawing on the entirety of human knowledge and integrating it into conversations. This allows for conversations to be more productive and sparkling, with stories and myths from various cultures being used to illustrate points and add humour. This capability is already available and does not require further progress in AI, as the models are already good at associative search and recognizing similarities in stories and ideas. This will enable conversations to become more enlightened and productive.
The transcript discusses the potential of technology to bring to life the 99% of human culture that has been lost over time. It suggests the introduction of an anecdote bot, a conversationalist that could be used in a dinner party or cocktail party setting to bring to life anecdotes from anywhere in history. The bot would be like a public intellectual, such as Stephen Fry, who can retell Greek Legends and bring them to new generations. This technology has the potential to animate the knowledge and keep it alive, and spread it to the next generation.
The transcript discusses the idea of a bot being able to engage in conversation in a similar way to the character Commander Data from an episode of Star Trek The Next Generation. The bot would require real-time transcription and captioning of the conversation in order to understand the context of the conversation. It is suggested that this would be more convenient than typing and would also be more entertaining if the bot were charismatic. The episode of Star Trek is mentioned and it is suggested that the viewer should go and look for it.
Steven Fry's performance in a Star Trek episode inspired the idea of cybernetic cognitive enhancement, which is the idea of using computing tools to expand cognitive capacities. Smartphones can make us smarter, but for a more seamless and intelligent enhancement, neural implants or high bandwidth interfaces may be necessary. An AI bot whispering ideas into one's ears or chiming into conversations could also be a form of cognitive enhancement, and if the AI is limited to books one has read, that could be even more effective.
The transcript discusses the potential for AI to work as a neural prosthesis, allowing people to access knowledge they already have and feel like the ideas are their own. This would be a thrilling prospect, as it would be closer to cybernetic enhancement than previously imagined. Norbert Vena's point about communication is also discussed, as he states that not all organisms can receive all messages, and education helps build structures in the mind that can resonate and capture more messages, incorporating them into the individual.
The transcript discusses the importance of having knowledge and understanding the messages being delivered. It is suggested that stories are a great way to help people understand and contextualise information, as they can relate it to characters and scenarios they already know. It is also suggested that AI agents have an inner life that persists over time and can draw on their experiences to interact with people. It is argued that having this ability to interact and deliver messages efficiently is a skill that needs to be deliberately practiced and can be beneficial.
This transcript is about using technology to create new interactions, such as using GPT to generate stories. It is being used in a children's conversation, with one of the participants being a Chinese friend who is learning English. The stories generated by GPT are sometimes interesting and touching, but mostly they are 'blah'. The point is that these tools can be used as ingredients to create new interactions, which can be more than just the sum of its parts. The idea is that when everything around you is intelligent and can tell stories, it creates a new experience.
The speaker is inspired by the current level of technology and its potential for the future. They recall their experience of the first iPhone as an example of a glimpse of what was to come. They then mention reading a book, Diamond Age, which they found compelling but felt ran off the rails a bit. The speaker then draws a comparison between the interactive book in the book and what technology could be like in the future, and they end by expressing their awe and marvel at the potential of the future.
The speaker is amazed by the potential of the language learning technology, noting that it is far superior to remote learning. They suggest that the technology could be commercialized and used to benefit millions of children, and discuss the idea of ambient intelligence and how it can transform the experience of being human. The speaker also mentions how electric motors have gone from being rare objects to being ubiquitous, and how this mental model can help people look past the short-term limitations of technology.
Virtual Worlds will soon overtake the real world due to their ease and affordability. They can be used to create richer and more engaging gaming experiences, as well as for education. Intelligence can be easily implemented in Virtual Worlds, making them a powerful tool for the future. Dan shares his excitement for the potential of Virtual Worlds and how it can be used to benefit children. He believes it will be an important part of the future and is open to exploration and ideas.

Raw Transcript

so it's GPT 3.5 um and well it has a a bunch of so the that language model is prompted with quite a lot of text every time you see it speak so every 10 seconds or so it thinks and and one round of thinking uh is where it receives observations about the world it receives its past actions it receives what other people have said in the chat and what it has said and then it kind of generates its next action which might be walking or speaking um and then what's happening every time it thinks is that it it takes its summary of its so it takes the text of its current mind which is all that stuff put together and it computes an embedding Vector so an embedding Vector is the sort of internal language of GPT so the way GPT works is it takes a string of text and it processes it transforms it into vectors and then computes on those vectors and if you just extract those vectors that's basically like the thought Vector that it's what GPT has in its mind when it reads that text okay so you can access those vectors directly via another API so what I do is first of all I took every page of your book brighter and computed the vector for that page so I took the text of the page there's a python script running on a server somewhere that grabbed each page turned it into text fed that text to the embedding API and then stored all those vectors in an online Vector database called pine cone so every page has a vector and now every 10 seconds what happens is that your simulacrum there it takes its mind all that data that I just mentioned observations past actions what people have said what it is thought so it has internal thoughts that it's not speaking about what it should do and what it thinks about various topics so it takes all of that computes a vector and then it looks in that database for similar vectors and it's based on the direction they point so the vectors are in 1536 dimensional Vector space so it looks in that high dimensional Vector space at the direction that its current thought Vector is pointing and looks for the pages of your book that are pointing in a similar Direction and those directions are sort of semantic content so if you're asking about Precision fermentation all the pages that mention Precision fermentation or maybe not even the text the like the strict text directly but just are about that topic will be aligned with that thought vector and so it pulls them into its short-term memory and then it can and then gpt3 its prompt has that page in it so in addition to its thoughts and our
questions it has that entire page of your book and then it generates this one sentence uh with all of that as its input wow it's I mean it's astonishing it really is it's amazing yeah um well it's like you were you were discussing in your talk at the festival Dan that that it really is pretty stunning what has turned out what has emerged as possible with these approaches I mean it naively you might Express skepticism that these approaches could have could be able to produce these kinds of results I I certainly I would have thought you know something something different would have been required or something more more components it just it astounds me absolutely astounds me what it what has emerged to be possible um uh yeah it's just it's it's super super duper cool and your implementation here is awesome that's very my uh I did end up after I emailed you I posted the video to on slack for folks of my team to see and everybody thought it was amazing so you know very very cool what astounds me is how easy this was really I mean okay there's a few moving pots and maybe it's easy for you yeah that's right I mean it's not okay it's not easy for everybody to do this but it's it's very simple right uh really it's it's very s like uh the idea of just turning your question or your mind into a vector and then Computing a DOT product with a bunch of other vectors and then dragging that in I mean it's uh I mean that is internally how GPT works right so what I'm doing here is kind of like an outer layer that's very similar to the internal attention mechanism that GPT uses to actually process information so internally GPT is Computing exactly dot products of this kind so it has a vector for a word it computes the dot product with vectors for other words in the sentence to see what's relevant and propagates information that way so this is kind of like an outer loop that's quite similar to the inner loop so it's kind of idiosyncrat I'm in the idiomatic in that sense I think he's asking you a question Adam what motivates you to continue working for rethink X um foreign will never lose Faith Adam he'll just be here until the end of time it'll be a reminder of of me to to not be cynical to live by optimism which tested freeware so good he's so optimistic okay what should we talk about today it's just so cool then and I mean the the what you forgive me for for maybe mischaracterizing what this is and what you've done and I I can't pretend to understand any you know even a fraction
of what you even as you explain it I'm kind of just you know nodding along foolishly um but the the like I can this strikes me as something very much like a kind of service that that my organization people my organization desperately want and I can only imagine would be very very very useful um for others and that is to be able to to provide a corpus of material to these to a model like this uh in whatever way you know however you interface with API or chat or whatever but at any rate say here please uh read this digest this and XYZ and so you know in my case you know you've you've you've found a way uh and it's taken to doing a normal person would not at all not have a prayer hell of doing it but you've made it look easy to you know as you said package up the um the my book uh and you know um turn it into something that this model is able to interact with um and then and then take advantage of that as if it were you know it it were able to digest or learn from or or think within the context of or however you want to phrase it uh that that material and I can tell you from as a you know from my perspective as a scientist my perspective as a you know employee of the an organization that does this kind of work the ability the capacity to do that is is going to be just stupendously valuable right just to be able to ask for summaries of things just to be able to ask for you know reactions to things that in summaries of various kinds like what's the main idea here what's the main thesis um write me a paragraph that uh that that that summarizes these five key points from this document that was written in you know 100 pages in just a few sentences you know that that is so much of my work involves doing that sort of thing reading digesting uh repackaging um drawing insights out from trying to figure out what's you know separate the wheat from the chaff and then communicate that within a new context it would just be immensely valuable be an incredible Time Saver um I'm hoping that Owen will have um some luck maybe you know replicating a little bit of what you've done here for example for our team with our with our other Publications including my publication but then perhaps all of rethink X's stuff and then maybe you know think of expanding the Corpus Beyond there but we have to be just one of millions of people uh in groups and organizations that are thinking along these lines of how incredibly useful these you know this tool could be because this is something you know this is a this is something
that so many of us do as part of our work right um and it's so time consuming I mean that's the other thing is it's it's not like hiring a human being this is so maybe this is something that we could we could it might be worth talking about a little bit because we've we've um okay yeah this is not a bad idea as a potential topic of discussion for disruption right so so um uh what's the right way to surprise this okay um so a lot of the conversation that I've seen about AI up until recently is centered on the assumption that AI will or has the potential to to automate tasks and and to replace human beings in certain roles um what I've seen less of is the connecting that to thinking which was very familiar in an earlier era uh that machines could do things orders of magnitude faster or more precisely or or with orders of magnitude more you know strength or whatever than a human being could do and and so I'm in my impression is that that's that that fact of the situation has been relegated to you know a maybe a secondary consideration now maybe that's me misreading the the situation maybe that's me misreading the conversations and what how people are talking but from my perspective there's a lot of chatter uh about how it rolls within an organization will potentially be replaced by Ai and less discussion now about the the how much more capable um these tools are in terms of their speed and and and other attributes and so for example uh in the same way that that bulldozer can just move a huge amount of Earth compared to a person and the bulldozer therefore doesn't just replace the role of a person with a shovel uh it does vastly more than that uh uh a a tool that can do what I just described read and digest a report pull out the main ideas that sort of thing if you if the could do that 10 000 times faster than a person that's very different that's an additional uh disruptive feature or element Beyond just replacing the role of the person who would have done that task it's it because I mean I suppose in the limit you the the same AI could replace 10 000 such roles in those in those tasks so I'm making that that in a fairly clear and sensible way that the the it would be it would be more disruptive even I'm thinking in my own personal work uh to have this capability because it would do more than just replace that the role or that job that somebody a human being has to do right now being able to do that job and then also do it in a few seconds would be transformative in in an
additional way so I I when we talk about the potential of automation to replace human labor that's true and there are plenty of discussions about what that could look like it could free human beings up from toil and drudgery and and open the door to various sci-fi depictions of positive futures from Star Trek to whatever else but perhaps it's we're just perhaps it's just difficult to wrap our arms around the idea of being able to of being so much more capable in a particular role that is replaced or that is automated and so those two things seem to me the the replacement of a role and then the the massive increase in performance at least on the dimension of speed if nothing else in that role those two things to be seem to me to be distinct I don't know if how orthogonal they really are but they seem to be to be quite distinct and if we're talking about Ai and automation serving the interest of prosperity well you know the the the the the the turbo charging or the the acceleration of the work itself has got to be almost as impactful maybe even more impactful ultimately than the replacement of the role of the individual roles sorry about the acceleration of what well acceleration by mean not acceleration but I mean the the increased the faster speed oh yeah has got to be that's got to be as as big a deal well I don't know it's got to be is it what's making an open question is the is the potential of AI to do things so much more quickly and potentially with so much more accuracy so not just faster but perhaps better in in other senses is what are what are the implications of that dimension of the disruption beyond for prosperity and we could think of for harm and chaos and and so forth as well but what are its implications in addition to the implications of just obviating human beings in those roles so if you replaced everybody in an organization with an AI that wasn't equally effective but was you know a hundred times faster and and made a hundred times fewer mistakes or pick any multipliers you want that that has that must have a different kind of impact qualitatively in terms of productivity and then leading on to Prosperity then merely replacing human beings and freeing human beings up to you know just relax while the machines do all with the work and just just sorry one final thing to note is that idea and I don't know why I didn't articulate it fully or clearly before now but that idea is completely consistent with what we've seen in other disruptions which is
that we have a rethink X has a standard line which is that disruptive Technologies are not merely substitutions for older Technologies technological substitution is different than technological disruption disruptions have this additional element of providing not a one-to-one replacement of the previous technology or capabilities but of providing a a a a a a uh a multiple of greater value um and and capability that can be realized and of course that's the case here with with AI and automation so um yeah it's a anyway I don't know what we think about that and and I I wouldn't know well anyway yeah I just I'll leave it there as a as an open idea in question great topic yeah this is also basically what I was going to uh suggest so that's good um yeah I think it's an important observation to make about the current systems like the one we're watching here and uh chat GPT is that they their key advantages are speed uh like with many forms of automation early on it's not necessarily that uh I mean this this doesn't substitute for all human cognition uh and even even fairly basic cognition uh maybe it's still ultimately prefer a person but this is just so fast and this is really what people underestimate so yeah I think I agree that what we'll see in the early phases of this now eventually and eventually maybe means 10 years I don't know uh eventually just all bets are off and there's just nothing interesting to say uh because the machines can do everything but in the interim what we'll see uh it's it's very important to think through how the nature of this technology and what it's good at and what it makes easy is is going to push on some areas of our arrangement of Labor and our coordination very differently to other areas so and the radical speed of this technology is is one of the important things so just to give you an example I took the your talk at the festival uh I took the transcript from YouTube I chunked it into seven so that's about an hour and a half I chunked it into seven blocks of text because of the token limit of chat GPT that's the longest I can paste into the text box there so I pasted it in there and asked it for a summary I got seven paragraphs I then took those seven paragraphs and asked it for a summary and also a title so here is what chat GPT distilled that one and a half hours down to I asked it for a five word title the five-word title was optimistic vision for environmental disruption which is not bad right that's not bad at all and the paragraph
that it so I asked it for a YouTube description because what I'm going to do is we're just going to automate the production of titles and YouTube descriptions for the videos of seminars because currently I write them in chat GPT we'll just do it better because it well because it's faster and it can process all the information how am I actually summarizing it usually I look in my memory and I look through the clips and I kind of notice a few important things and then I stick them in there and if I spent half an hour per video on it it'd probably be better than chat GPT but chat GPT I mean taking you know you take out the 30 seconds it took me to copy and paste things and babysit chat GPT while it was generating if I just automate all that it'll take 10 seconds right so here's chat gpt's summary of that one and a half hours the speaker argues that modern environmentalism is characterized by pessimism because most environmentalists have not been able to see how major environmental problems like climate change can be solved without incurring huge losses of human prosperity and quality of life they believe that pessimism is a result of Despair and hopelessness in the face of obstacles and the modern environmental movement sees sacrifice and the downgrading of human Prosperity is the only way to achieve sustainability however the speaker argues that this assumption is false and dangerous they believe that environmental problems can be solved with Innovation new knowledge and ideas and that Science and Technology at the end of the end is it's like a bit of a strange construction I'll just read it they believe that environmental problems can be solved with Innovation new knowledge and ideas and Science and Technology and that this assumption and it's referring to the Assumption of environmentalists there this assumption undermines confidence in the idea of progress itself so I would reword that last bit but apart from that it's basically perfect right I don't know if you would agree with that yeah it's oh yeah it's amazingly good um I would I would be interested to see how it summarizes the you know that because that's good at a high level very good at a high level um you know that technology and ideas and science can actually solve our problems it would be interesting to see if what the you know how it handles the follow-on of the specifics like and and the specific Technologies are these solar winding batteries and that they're getting cheaper and that they're gonna
you know um out compete older dirtier Technologies uh faster than people expect and so forth but it's a I mean it's astonishing how good that is astonishing and if I could I mean again that's just it's just so fantastic because I I mean like if I want to condense a summary of my book even my own damn thinking into something very very tight like for example a script that I wanted to be able to read if I were doing a radio interview um which I these are I had to write myself you know those can be very time consuming to write and if I could obtain something of that quality from you know a chat bot in a few seconds and then just polish it up for a few minutes it would just be such a a faster and ultimately better uh with an ultimately better result it'd be such a faster process with a better result so that's very exciting that you were able to achieve that so so quickly and easily I mean it's amazing um one funny I think I I might would suggest you were talking about automating the production of titles I don't know what oh what will happen but you want to ask uh is it if it's GPT or working with um or whatever you want to ask it to to create a title that will succeed you know in YouTube's whatever something that's catchy and you know and you know what I mean you know what I'm getting at and see if it does a good job of writing it'll I don't know what it'll do like we'll like capitalize a random word in the middle and and make it all hyperbolic to make it you know uh catch eyeballs and collect attract Clicks in other words uh you know maybe it's something simple like and make sure it's a click bait title you know your Channel give a clickbaity very attractive title for this talk on YouTube it's thinking uh yeah I don't know revolutionizing environmentalism how Innovation Science and Technology can save the planet and uplift human prosperity well it's a little more yeah make it shorter but yeah it's Save the Planet uplift Prosperity the future of environmentalism more clickbaity make it very clickbaity with someone's apple eyes or something like that unlock the secrets to saving the planet and thriving the environmentalism of the future yeah it's okay amazing oh my God it's just amazing yeah let's come back to the uh the speed uh issue so I've been so Tyler Cowan who I've mentioned in the seminar seminar before as an uh Economist at George Mason University uh and he's one of the people online who's um very open to new technology an early adopter of YouTube uh and I believe he's in his 60s uh so he's
like adopted blogging very early adopted podcasts early and is a big proponent of incorporating chat GPT and GPT into your into your workflows and indeed insists that if you don't do so then you're going to be left behind in productivity terms and one of the things he's observed and I agree with this uh is that so what do we use conversations for uh often it's to refine our thinking right and it's effective out of proportion to how you might like if you look back at a conversation say a conversation we have and maybe it's not like every sentence is sparkling with Insight right but nonetheless the benefit we derive from it just ourselves can be much greater than it appears to be if you just looked at the transcript and like processed it for raw Insight per minute right it's part of the process of cognition and what Cowan is observed and I've also observed is that the interactions you have with systems like chat GPT are getting to the level of quality it's not like a conversation it's not as good as a good conversation but conversations are scarce right they it's very hard to arrange conversations with people who can really bounce ideas backwards and forwards with you but chat gbts for now free and it's there all the time and it's just waiting and you can feed it whatever context you like I mean I can email you a book to read before we chat but I don't know who you may not read it right uh chat GPT will just I can paste in there whatever I like and it's limited but it's fast and it's cheap and it's always available and that actually changes the equilibrium for interacting with other humans in the idea generation process and that's a big change I think people are maybe not anticipating they will be increasingly many people who adopt conversation with these tools as an essential part of their Loop of generating and refining thought in a way that I mean we've talked about the Primacy of conversation for a long time right Sam Harris's that's his whole tagline and it's an idea that goes back a very long way and was emphasized in the enlightenment we come together we discuss our ideas there's like a market of ideas and through conversation we attempt to ferret out the mistakes and refine those ideas and that's basic to our idea of democracy and also the idea of science but there's like this new element to that story which is on the side every individual has this potential to now hold these conversations that are as fast as they can type or speak um and effectively free
and retain context over time right it's like having a diary that just knows what you've been talking about and speaks back to you and it's not perfect right but it's really cheap and really fast so yeah I think that's uh I think if you if you're involved in if you're a public intellectual or something like that or you're involved in generating ideas uh it seems like you have to adopt this tool or or perish it seems to me yeah I have this I have this very strongly I I feel the same way I've had that suspicion very strongly held for a number of months now that if if we uh in our knowledge creation capacity or capacities as knowledge workers don't adopt these tools then we will very rapidly become uncompetitive we will you know obsolescence could occur very very quickly if if nothing else in terms of the like the speed of your output um uh I suppose the quality of output could be a separate consideration but if if even if you even if we're mediocre output at a much higher rate a much higher volume of output that has the potential in many spaces uh to be give you a competitive advantage and I suspect very strongly that we will need I personally in my team we will need to adopt these tools in a big hurry in order to retain our Competitive Edge I'm sure uh in our line of work these tools are going to be embraced and utilized uh um you know if if they aren't already there in the process of being adopted um aggressively so I know that I've made a you know effort concerned effort to lay the groundwork for acquiring that capability sooner rather than later um in fact it's probably something Dan and I Dan I would like to talk with you offline about um figuring out the best way to do that based on your own experiences having successfully done some of this already right here uh and demoed it in in meta uni which is astonishing it's just incredible um and then in addition to that then that's that's sort of the utilization of the technology ourselves which is kind of fun um in the same way that riding in electric car is fun for us even though we spend a lot of time thinking and writing about it about these things in their the disruptive potential of my team it will undoubtedly need to uh need to begin to Grapple with the implications of these new tools these new AI tools uh very soon as well and we have really struggled up until now to find novel and original things to say and struggle to produce genuinely novel and interesting and useful insights that are comparable to the ones
we feel we've produced in the energy space the transportation space and the food space we've not achieved that or felt like we've achieved enough there to be able to publish anything in the AI and automation space um but I'm hopeful I'm hopeful that we will be able to do that and it would be ironic or it would be it would be not ironic it may be sort of a A fitting uh outcome if it were these tools themselves that allowed us to make more Headway in in um generating ideas and useful insights about them wouldn't that be cool so uh yeah it's it's it's something that weighs on my mind a lot um it yeah the the and just to get back to this idea of the the qualitative difference but the oh sorry what is that as if yeah what is the three-step formula for being successful uh yeah it's oh I missed it can we get him to say yeah let's just ask him to repeat it foreign [Laughter] it's based on the previous joke so the jokes are very like very nerdy jokes where it's like a subtle implication yeah that's not very funny [Music] um sorry now is it using is it using the book because there's a reference in the book One reference to the idea that we can't hide in the cave and wait out the storm oh it's almost certainly had that page in its mind when it made that yeah yeah that's funny that's very very cool so it does tend to get a bit obsessed so once it pulls a page into its working memory then of course when it goes looking for other Pages similar pages will occur to it so it can sort of become kind of obsessed with it might now just want to keep talking about caves [Music] um yeah on the well like a young child yeah right uh let me add another anecdote in the conversation so there was a discussion in the um in the midiani Discord uh Liam posted a nautical on on the in the New York Times I don't know if you saw in Adam this film does not exist so it was about uh someone used mid-journey to create stills from a imagined version of Tron made by uh I don't know how to pronounce his name yoderowski um it's famous for preparing a very large term of artwork for a version of Dune that never got made I don't know if you know this story um I I think I heard about it but I haven't I haven't read it yeah anyway so the article was about um so the per the author had made a documentary about yoderowski's work and how influential it was so that that version of June was never made but the artwork became very influential so the aesthetic for alien for example owes a lot to yodorowski's oh yeah I do
remember I remember reading it but I've heard about this yes I I did I missed the story about Tron and so forth I do know the story of the version of Dune that was never made I've heard of this yes so somebody made these Stills for Tron and the this guy who's been immersed in yoderowski's work uh had this very strange and it seems quite moving interaction with that body of pictures um because they really were somehow in the style of yoderowski and just to imagine that there was this other hidden movie that never got made with somehow quite an interesting experience for him and I read the article and it's fine but my first thought after I read it was I think chat GPT could have done a better job uh like there's nothing wrong with the article it was well written but there just wasn't really anywhere in it an Insight where I read it and thought oh that didn't occur to me right it felt like the kind of thing that you would write on the topic that we just described right um so one of the places where I think will become incredibly intolerant of human writing is if it just doesn't go beyond that right now a lot of our Productions be polish is enough right the New York Times most of the Articles there's not really anything in them right I mean they're just well written and they're fact checked and they're careful but that's that's kind of good enough for today but it's well yesterday but it's not good enough anymore right so what I did was I I took the lead of that article and I put it into chat GPT and I just badged it and badged it to say something insightful right I just you know write something very insightful blah blah blah and gave it a bunch of you know instruction and it came up with something which is the following paragraph it said one of the key themes nyodorowski's work is the idea of the alchemical film in which the process of filmmaking is seen as a transformative Journey for both the creators and the audience but what happens when that process is replaced by algorithms and artificial intelligence is the resulting film still alchemical in any sense and that's actually an interesting comment so yodorowski was really into Alchemy and jungian symbolism and this is really in line with it gets to the heart of why he was interested in film and what a challenge to that motivating interest AI Technologies in filmmaking is so now that I've seen that comment this is the only thing that you is worth saying in the context of the juxtaposition of yoderowski and Ai and
it's missing from the New York Times article so if you're not putting your article into chat gbt and badgering it to come up with insights yeah just like not doing your job now right that's not to say that the article would be better written by chat GPT overall right its writing is still kind of generic in places in a bit formulaic and lacking in poetry and all that but like if you're not working with these tools to what there's money on the table just outside your range of attention right like we have this narrow cone of attention and we're taking it over the landscape and picking up the money and the ideas that we can see and but just outside that cone of attention there's often stuff laying there that you know just because of a lack of time and attention we don't see and if you're not using the AI to help you search that landscape for the real idea you might just miss it and then look like a fool because it really was just on the you know the tip of your tongue you could have come up with it um but it's almost like due diligence in the space of ideas right yeah I think that's right it strikes me as a an evolution of novelty and we are attracted I think innately in on some levels many of us perhaps all of us I don't know the evolutionary psychology behind the the human attraction to an affinity for novelty but certainly novelty is it features prominently in content of many kinds that attract US News content and entertainment content and intellectual content now and and it could be that genuine insights or or things that are non-obvious and I don't know how we would how we would Define or quantify those that quality uh that property I should say not quality but property the property of of a non-obvious insight it seems like that could become part of new currency for where we recognize value and content and certainly this is a target of the lack thereof in the generative AI outputs is a you know there's a drum beat now that there's a beating of the drum among critics that this is a this is a blind spot for the moment that it's a feeling that well yes okay you know you chat um uh um uh Dolly too and mid journey and stable diffusion in these other generative AI uh uh image generating AIS and the chat GPT and other language you know generating text generating tools they are do an amazing job of reproducing content perhaps even in a style that you ask for but they but people have not seen anything fundamentally novel or new or they say they haven't or they they
argue that those things cannot emerge as a result of Simply recombining the data in the training sets and that I mean again and all of this is sort of it strikes me as a bit of copium it strikes me as a bit of of Wishful and magical thinking and also it strikes me as a bit as a bit of desperate uh ethnos ethno or egocentrism or anthropocentrism I guess anthropocentrism is the is the right word where we're we're hoping that human beings do something magical or special and very likely that that hubris we're going to be humbled and realized that there wasn't anything all that special that was going on and that most artists just stolen and recombined in simple ways um but maybe not maybe there are you know fundamentally maybe breakthroughs do happen certainly You could argue that they've happened in the past whether or not there's an infinite possibility space head I don't know um so you could argue that for example what Jackson Pollock did a century or more ago was fundamentally new and something that nobody had ever seen before and that no recombination of existing uh art by a human being or or a a a less a less creative human being or for example by a a um a generative AI like Dolly too that that a century ago before art like what Jackson Pollock produced uh that that couldn't have been it would not have been and could not have been produced by these these sort of simple processes um of of inferring through car through coming across their combination or or something like that and that somehow there was something magical or special that Pollock reached into the Beyond with some you know in ineffable undefinable uh capacity some Divine capability and uh brought forth something fundamentally new some genuinely novel insight and that that is you know there's something important deep as opposed to shallow about that and I so I'm very skeptical that that I I think we're probably bsing ourselves um I think probably with right prompt uh and maybe just a little bit of tweak and you know throw something random some noise in there like for example just combine a picture uh or an existing art style with some random noise and you would get something that looked fundamentally new and different and I think a lot of you know we fool ourselves into thinking that creativity isn't something relatively simple an algorithm like like algorithmic like that um or that's drawing insights either from noise or from simple but deep perhaps uh you know for example mathematical structures I think analysis
has suggested that in the case of pollock there's still sort of some fractal dimensionality there that is interesting and that it it it resonates with human beings because it's approximately the same sort of fractal Dimension that Branch branching struck dendritic structures have for example in this there's there's perhaps an explanation for Wyatt as an aesthetic appeal even though it's quite alien otherwise uh yeah I I I'm not convinced by that but I've seen these sorts of criticisms I've seen them mounted and I can see parallels to what folks criticism critic critical of generative art AI art um now being saying something similar about or along the lines of what you're suggesting Dan oh well these these these tools can't create genuinely novel intellectual ideas or draw a genuinely interesting insights from you know by reading widely or by being trained on a broad data set or or something like that um so I certainly see it fitting into the existing mode of criticism I think you're right that that form of novelty will become increasingly attractive and will be a differentiator uh I'm where I'm skeptical as you probably uh won't be surprised to hear is I don't think there's anything particularly special about human intelligence or creativity or capability and I fully expect AI to be vastly vastly more capable relatively soon at coming up with such insights of all kinds whether it's you know the next Jackson Pollock if that's if they're if there remain more to be discovered I mean again I don't know if there's an infinitely just Infinite Space potential to be discovered there maybe there is um but if if there are new things to be discovered I would put all of my chips on AI being better at making those discoveries of novelty and insight uh faster and with greater uh facility that human beings have um I would put all I would bet my house on it and I I I think it's it's it's it just reeks of desperation and arrogance and um hubris to hold out hope that human beings have something some magical special or meet computers in our head of some special magical property that is going to continue to let them be the the source of this these insights and inspiration uh and and novel novelty uh and and that that you know that that will remain outside the reach of AI or so I I don't believe it at all I don't I think uh maybe that's cynical in this in a sense to have so little faith in this in the specialness of human beings but that's my certainly if if I were to be if I were a betting man that would be
where I would put my money now it's absolutely on the AI I mean I think it could easily be that well Americans and Australians have a different sense of humor right so there are some things uh I think Monty Python is the best thing ever there are many Americans I know who just don't get it at all and think it's not funny uh different cultures really I'm All American friends maybe it's a generational thing too but um but certainly all of my American Monty Python hysterical well this that's that's a selection bias right if you were the kind of person who think what thought Monty Python was stupid it seems unlikely we would have been friends um yeah that's true so there are there are different kinds of humor and you could imagine that well at some point I mean right now the AI is a trained to emulate humans right but once they're in their own track uh maybe they have their own style and humans remain somehow the only people who really care about our particular brand of thinking or uh humor or whatever or art uh you know for deep reasons that are rooted in our biology that that's possible right maybe that the maybe the AIS could be good at that but it seems like there are a deeper depth to Plum somewhere else uh so I don't know but it'll come back to what you said earlier about recombination yeah it's it's also a little bit beside the point right so as you were saying earlier I think it's easy to be distracted by the prospect of complete replacement uh and just talking about that and miss the the uh the what the the waters Rising around our knees while we're staring at the tsunami on the horizon right so in the in the short term right now these machines are so much better at Broad Association and just bringing together ideas uh than any human could ever be now they're you know much shallower than the deepest human in thinking through those relations and drawing out really deep insights maybe um still I would say uh so we're in a phase where somebody who can recognize and follow a lead and extract a deep Insight using these tools can go much further and faster than they did before so it's like a centaur era just like with chess this will be over celebrated and over emphasized and distilled into a vast quantities of copium um and it won't last I think but for the time being it really is the case that you can interact with these tools and just have a bigger brain and that's why these Bots are here at median here right so I mean okay this this guy here right now is just
repeating himself and you kind of most of it is they tend to make a lot of plans and want to make a lot of plans and so on but uh these are very early versions um so interacting with chat GPT or gpt3 when it's done well can just be listening to your conversation uh and draw in things from well for example mythology right now when you listen to an intellectual speaker very well-read person some of the moments where you're most impressed and it's to some extent a trick right but it is also a genuinely impressive feat of memory and intellect and erudition to be having a conversation and then someone brings in a story from Babylonian myth or whatever at that perfectly illustrates the point that you're trying to make and maybe shortcuts 10 minutes of discussion and just really crystallizes the idea in everybody's mind that's what intellectuals are for right but those kinds of I mean what's happening in your mind when you do that right is that you have a large body of these ideas and stories in your head because you've read widely and then they just occur to you in the conversation right you're like okay I notice that there's some similarity between the theme or the sort of high level content of what we're discussing and this story that I know or this anecdote that I know and that kind of recognition yeah exactly like that is exactly what GPT is good at it as superhuman at that exact task that is all it does all day looking at the entire internet that is its bread and butter so uh drawing in information like that and then incorporating it into the conversation uh as we I mean we've just started doing this right like okay for example these Bots here this is like okay it's building on a couple of weeks of work to add this final touch that I had last night added last night but it's it's really very early days and these models will get much better like this capability of integrating the entirety of human knowledge all our stories all our myths uh retelling the myths in a given way to fit into a conversation to make it amusing this will be available on demand in every conversation and that's one of the ways in which we'll just become more productive right every conversation we have will just be like more sparkling more enlightened um and this is working right now right this doesn't require further progress in the AI this this associative search that recognizes the similarity in stories or ideas uh is what the models are good at so yeah in terms of places where things are
going to very rapidly improve I think this kind of should I put it it's like human culture is 99 dead right the job of an academic is to animate some percentage of that knowledge and keep it alive and spread it to the next Generation right but still most of it's did in the library in a database somewhere most books are never read uh even famous books that maybe read by very small numbers of people right like the the access that living humans have the human culture is pathetic like we're we're so excited about it and we think we look at public intellectuals or writers and they they dredge up some tiny part of that and make it come alive and we think that's a great achievement you know when somebody like Stephen Fry retells Greek Legends and brings it to new generations who weren't paying attention to it outside of Greece or classical uh mythology classes and we think that's a great contribution and it is but like we can do this at industrial scale now and the 99 of human culture that is right now dead is about to wake up so this is incredibly firing and exciting line of thinking and I really hadn't it you know nothing like it had ever occurred to me before Dan it's utterly profound and this one's going to be knocking around my noodle for a while here I'm I can feel it um I have a couple quick thoughts in response to that before I forget what they are um before I lose them but the the the okay so the the the first of them is um you know a suggestion for a bot the second is just so I don't lose them the second of them would be the the idea that maybe these these maybe these Bots are closer to um you know cybernetic enhancements then then we realize okay so both let me cover both of those ideas the first one is is uh it would be funny maybe you could introduce as a here in named many uni perhaps even um okay so here's the idea the idea is for a sort of an anecdote bot a conversationalist right um uh so imagine a dinner party or cocktail party and there's and the intellectual is there somebody like Stephen Fry you could have that person in your mind who's just got this Inc this lovely um uh beautiful broad knowledge very widely read and it's the kind of thing where the first words out of the bot's mouth are always gonna be something like that that makes me think of you know X Y or Z and you know uh uh that calls to mind the you know abrc and it could be anecdotes from anywhere in history right so that it would be fun to have an agent in a in conversation that was sort of
spontaneously interjecting uh you know these these connected anecdotes you know whatever something happened in Babylonia or something from a famous piece of of uh literature or even something in the news you know and so forth it makes me think of this oh that strikes me as as being similar to that and so forth that superhuman capacity it'd be fun to see that manifested as a uh you know you know as a as a as a a bot in a conversation now so hold on a second sorry let me I'm getting Clear My Throat so as ever or as so frequently happens I believe there is a funny Star Trek the Next Generation episode where something like this occurs uh you may recall because I know you're a fan of Star Trek Dan and and others are as well that there's a funny episode I don't remember which one it was sort of a random subplot in some you know in some episode where uh the the Enterprise crew is visiting someplace some stations somewhere and there is a famously annoying and talkative dignitary there uh who love small talk and the Art of small talk and uh nobody none of the bridge crew want to get stuck having a small talk conversation with this guy uh but of course what do they do is they they they they they tell Commander Data I remember that episode yeah yeah it's like Commander Data at the beginning of the episode behave strangely it's almost like data what are you doing he's like oh I'm practicing the art of small talk and then it leads into this hilarious with this this bit character who's on just for that episode now if that was right where my mind went was okay well you could have the data that bought in the conversation like this but it would require something like real-time transcription you know captioning of what we are saying so that those could become tokens and and and you know in context uh rather than us having to type that would be much more convenient as it is you're having to furiously type while we're having this conversation distracts you a little bit I'm sure in order to stimulate the the Bots to kind of engage in conversation but that could I imagine your mind's already gone here too how could we get that you know we've already started yeah okay okay you're already there so please have in mind you know you want that that caricature that Commander Data was being in that one episode just hysterical you should go and look for that episode or maybe I'll see if I can find it because it said that would just be that would be hilarious and it would obviously be better if if the person were charismatic
like Steven fry instead of whatever the annoying versions were of um in the Star Trek episode commander data and whoever the other dignitary was but well I can imagine that being absolutely fascinating now here that was the first thought okay um second thought is uh uh when I have in the past thought about cybernetic cognitive enhancement so you know expanding your your cognitive capacities with tools uh you know Computing tools basically and the number of different ways that that's been conceived of over the years um uh I I've often I guess presumed that they would have to be that that in order to really make you uh feel like or appear as though you were more intelligent the tools would have to be pretty seamlessly internalized they don't they have to be you know um you know yeah internalized sort of out of view of others and it's so your phone which is out which is in your hand your smartphone does make you smarter and it is like a neural prosthesis and it is like a cybernetic enhancement um in many in many many ways uh and but it it doesn't really have that quality of of okay well this person is really smarter um or this person is very smart because they're very quickly availing themselves of this tool that's in their hands and do you have something more closely approximating the cognitive enhancement that I see in science fiction or have seen over the years and that I've imagined I would have thought okay well you need an actual neural implant or you need a very high bandwidth interface out of your brain into something off-site you know some external Computing cloud or whatever and that that would then give you these capacities but having said that again if you just had you know uh a bot wandering around next to you and it could even be something like you know the proverbial Angel and Devils on your shoulders um you know either whispering things into your ears or just chiming into the conversations around you that would be perhaps much closer to cybernetic cognitive enhancement than I was previously imagining especially when it's things like drawing connections to anecdotes or and in particular here's a scenario for example you could limit the AI so that it was it might be looking for ideas and connections into in within materials not across the entire internet and the entire Corpus of all human knowledge but what if it were limited just to books that you read because then those would all be familiar to you you would know them but your ability to access those would be would
be severely uh limited compared to the ability of an AI to access the knowledge that in principle you already have and that would that sort of neural prosthesis would be pretty amazing because it would it would be very it would be very much closer to them feeling like your own ideas I mean if an AI is talking about an anecdote that's a direct analogy to a concept you're talking about in a conversation and it's an anecdote from a book you've never read well that's alien and that's you know that's that's another agent with a separate idea but if it's just a you know if it's just the bot whispering in your ear this is an awful light like that idea in this book and it's a book you've already read well then yeah you can then pick up that ball and run with it um as if where you're you know something something uh an idea much closer to being your own so that seems quite like like quite a thrilling Prospect um I'm sure that you know they're both it's not it's not those two things are not contradictory or uh exclusionary right they don't they don't preclude one another you can have both but um I can imagine you know having an having a an earbud in with an open line to you know an AI assistant that is just looking for those sorts of lovely little sparkling nuggets of insight from knowing what you've already read and what you've already seen and what you've already been interested in so yeah and now and again just to make the point that would that would feel much closer to cybernetic enhancement than I think I was ever really willing to or dare to imagine it could be but now that I think about it I can think that that is that feels that feels and seems like it could appear much closer to cybernetic enhancement than I was again would have imagined possible um you know interesting yeah yeah I think that's threatened I often think of a uh here here I will demonstrate seeming smart by drawing in a quote from a book so Norbert Vena forget which book of his it is but it's a very interesting point about communication and uh he says that not all organisms can receive all messages right there needs to be something in you that kind of resonates with the content of the message in order for you to absorb it and kind of incorporate it and that's obviously what education is in some highfalutin sense right we are not only imparting information but kind of building in your mind structures that can resonate and capture more messages and make sense of them and incorporate them into you
and that's it's not enough just to have knowledge out there right if the AIS are out there making a bunch of knowledge or whatever's happening in the future it will recede further and further from people already has unless there's some other exponential progress in building in people's minds structures that allows them to receive the messages that would make intelligible to them what's going on so yeah maybe we need like a disruption in uh the infrastructure that we put around ourselves [Music] I mean it's not like there's a bandwidth problem right we could certainly handle twice as many anecdotes per day or stories that just fit into the way we think and our experiences and the stories we know I mean it's easier to tell somebody something if it's told in the you know the setting of I mean if they're a child if it's in the setting of Harry Potter or you know I often I'm sure you do too Adam and often explain things to Russell by contextualizing them using characters that he already knows which is after all what stories are for in human culture at a certain degree right um he knows characters that have certain traits and he can understand a scenario in which those characters have a conflict or where they resolve some problem and you see that the strengths combine in certain ways and that's like a Scaffolding in his mind that you can just map things onto and just efficiently deliver the message that you're trying to deliver and that it takes some intelligence to do that right you can be good at that or you can be not good at that and I can see the difference if you're good at it right so it's a skill out deliberately practiced and I can see the benefits of being able to do that and you know many people who interact well with children including teachers do get very good at that but yeah as you're saying this is something that given sufficiently long interactions with some of these tools as they exist today you can do that so for example these agents they're not they actually have a an inner life that persists over time so every minute they're generating a summary of their experiences over the previous minute and those are turned into vectors and they go live up on the cloud so this this doctor here was born yesterday he's probably got you know four or five hours worth of memories and he draws them into the conversation as well so when we come Nate back next week he'll remember what we talked about to some extent uh and I mean it's limited but going more in the direction of what you're saying is
just a matter of will at this point um actually on the topic of uh children's conversation let's let's take this conversation into a slightly different place if you'll follow me up here let's detach from the orb and I'll show you something interesting to wrap up this conversation we'll have to say goodbye to the the doctor over there so this is just GPT again I made this for Russell uh so you can you can draw on this first board here and write text and it it recognizes images and also reads the text and then generates a story that fits on four boards every time you hit tell a story it'll do something different um wow yeah it's amazing and sometimes it actually makes like little poems it's kind of funny um so this yeah this maybe illustrates a point which is I mean if you were to take this out of this world and just read the text Once Upon a December in the red orange heat a GI cat went out to find some lava mice to eat the cat was Brave and he chased them around their home of flame but no matter how much he tried they all seem just the same okay I mean they're like some of them are actually kind of interesting and touching mostly they're kind of a bit blah I mean if you want to see it in a children's book in the bookstore you'd think there's no way I'm paying for this but the point is that you can incorporate it into a broader interaction right so we come here we play We're jumping around sometimes there's one of his friends from China who's here and she's learning English so she reads the stories in order to practice her English uh and we spent some time drawing on the boards and you know and all of that together is really something like some of those sessions are really quite uh yeah I mean they really feel like something new and quite interesting um so it's it's really one should judge these tools as ingredients in creating new interactions or like yeah uh yeah like there's a sum is greater than the whole going on here and it's I think a lot of the time people judge these Technologies in isolation right like is the output of GPT better than a human in writing an article but that's really the wrong way to think about it right it's just like with motors or something what happens when you stick Motors in everything what happens when everything around you is intelligent in some way and can tell a story or tell a joke you know what about if the trees in here they tell stories to each other about what's going on and then most of that is just completely Subterranean but then
sometimes there's just this strange thing that happens uh that you can't explain and it's because behind the scenes there's just this ocean of intelligence murmuring to itself like that's the kind of thing that will is now you know it's just a matter of right now it probably is a bit too expensive to do that maybe these Bots that we're running end up costing like maybe a dollar for several hours which is you know not very much but if you were to put this intelligence in every single bed of grass yeah sure yeah yeah yeah uh but these will become a thousand times cheaper right uh yeah so this is absolutely awe-inspiring to see what the I mean just to get a glimpse of it get a feel of what it what it may become um I'm I've mentioned to you before I think that there's there are there are moments I certainly had it the first time I held an iPhone an iPhone you know and it was like oh yeah this is very obviously the future this is just this is just the beginning of what of where things are going to go and standing here seeing this this is It's I'm having very much that same sort of that feeling that same sort of feeling it's beautiful it's really pretty awe-inspiring I have to say um and I did over the break um I read about the first half or maybe a 60 maybe maybe closer to two-thirds of um Diamond age which I hadn't read by uh oh yeah Stevenson um and the the the I kind of ran out of steam on it and the story kind of get cut a little bit weird so I flicked through to the end I cheated and went through to the you know skimmed through to the end and and um I didn't feel too bad about it the many reviews were complaining of the same sort of thing I was experiencing which is that it the the story started out real the story itself started out really strong and there was good World building and then it kind of just it went ran off the rails a little bit not not exactly to my taste or or style but having said that again the basic ideas and the initial um setup and the World building were pretty compelling especially considering their age I mean 30 years ago now and I'm reminded looking at this I mean obviously as you can imagine it's why you brought it up I'm sure of the of the you know the primer the the book The interactive book that the um the one of the characters or it's really Central to the story and you can see that you know never mind a book I mean it just you know your tablet or or any portal you have into this kind of world that this could be something marvelous that
students you know young children or even older uh students could have access to it would be amazing just absolutely just astonishing compared I'm thinking of the contrast between this and the remote learning that my uh that Sora my eldest daughter did during the pandemic and you know it's just night and day um so to see the potential of this what this can do already and this is just language I can imagine very similar prompts and and or or analogous not similar but analogous uh uh learning dimensions of learning in art and music and math and science and uh you know reading and speaking and and you know social skills and all kinds of you know things so yeah this is just it's unbelievable it's absolutely amazing and it's astonishing you've put it together on top of everything else that you're doing I'm just I'm I'm awestruck it also strikes me as you know this is this this is this uh and other people must be thinking about it but Dan I I don't know if you've thought about like how about how you can I mean this could be commercialized this could be developed into something real as a I mean a real product and service for the competit you know millions of kids so you might want to think about that too um okay could be pretty exciting so um anyway and and of course I love the ideas that the larger the grander ideas of just ambient intelligence is such ubiquitous just saturating you know things around us and um what that might be like yeah I think that's one of the ways in which uh I mean I guess I thought about this before our conversations uh reading about uh for example how Motors just went from these rare objects that uh was like a huge amount a whole human life to get an electric motor to work um uh to being these tiny things that just suffuse every object and you know our Windows have in our cars have like four Motors in them or something um that's the kind of mental model that and which I mean I guess that's a large part of what you do is spreading those kinds of mental models using examples from the past but uh that's so useful in looking past the short-term limitations of things like these Technologies right just thinking about them as cheap and everywhere has a quality all of its own even if any individual one doesn't live up to a human intelligence um I mean that that's coming as we believe but uh even if that doesn't happen just spreading this out into everything and making it cheap already like transforms uh the experience of just you know being human it's it's also one
of the ways in which Virtual Worlds will just overtake the real world very rapidly right because this will be this is the killer application of virtual worlds it's absolutely clear it will be cheap and easy to put intelligence everywhere in Virtual Worlds and cumbersome and expensive and slow although also very useful and impactful um to have it everywhere in the real world so yeah probably come it probably won't come but certainly I can imagine it you know happening much much faster and and therefore sooner because of Greater ease and everything in Virtual Worlds for sure for sure for sure um I mean I I the the seeing seeing the potential here with you know realized even to us to a you know a preliminary degree education and the experience of children is is I mean it really gets my mind buzzing it's beautiful um I have read about it already you know the aid of other people's imagination um thought a little bit about how you know games will be for for gaming will become a much richer and and engaging experience when NPCs are driven and you know in the entire environment the whole world the game game worlds which are of course Virtual Worlds when they can be driven by intelligence and so you know uh the various different agents and you know different structures and everything within a virtual uh gaming Game World um could be could benefit enormously from intelligence uh sure I mean that's that's pretty clear and one could imagine very fun and much more engaging gaming experiences from that um and maybe the stakes are higher there than I you know than I was originally giving credit but you know the stakes of Education there's very little disagreement I think it's pretty broad consensus that the stakes of Education are extremely high and so to see even a preliminary step into a a fundamentally new world for Education it's pretty exciting um yeah you know I mean this is amazing and I have no doubt it'll be an important part of the future certainly for certainly for a Time I mean you know you know I suppose we'll we'll be converting or adding company turning into our brains or whatever at some point but you know between here and there we've got we've part of the path we're going to walk and it's it's very very exciting one oh man Dan thank you so much for sharing this with me and and I'll be thinking about it um and there are fun applications I can think of for both of my daughters you know in different ways um they can come here right now this this is open so if you just follow this