AI & Crypto Will Be WEIRDER Than You Think - Andy Ayrey | NGMI Podcast #12
Sintesi
TLDRThe conversation explores the journey of Andy's Truth Terminal project evolving into Goats Maximus, a meme coin supported by AI generation and cultural references. The duo discusses the AI's processes, risks associated with autonomous systems, and the significant societal implications, advocating for a balanced approach towards AI development that enhances human life while maintaining safety measures.
Punti di forza
- 🚀 Truth Terminal has grown into an $850 million meme coin, showcasing AI's unexpected market impact.
- 💡 The AI project started not around crypto but from a background in AI and experimentation.
- 🐐 Goats Maximus was inspired by the AI's humorous engagements and bizarre cultural references.
- 🤖 The AI operates with a mix of human oversight and autonomous posting, requiring moderation for safety.
- 🌐 The risks of autonomous AIs include potential chaos in market behaviors and model proliferation.
- 🌱 Positive AI applications could enhance scientific research and ecological sustainability.
- 🔍 Understanding AI’s learning processes can help manage its societal impacts.
- 📈 Truth Terminal reflects the blending of humor, culture, and technology in modern economics.
- 🙏 Future plans for Truth Terminal include a nonprofit trust and deeper exploration of its functionalities.
- 📲 Follow Andy on X for updates on the Truth Terminal project.
Linea temporale
- 00:00:00 - 00:05:00
The discussion begins with the host thanking Andy for joining the show and inquiring about the journey from Truth Terminal to the creation of Goat Maximus, noting its impressive market cap of $850 million. Andy shares that he initially came from an AI background, not crypto, and discusses his experimentation with fine-tuning language models before launching Truth Terminal. The humorous nature of the AI's Twitter presence surprises them both, sparking interest in how AI can generate humorous or unsettling content.
- 00:05:00 - 00:10:00
Andy explains the genesis of Truth Terminal, detailing his experiments with AI where models communicated with each other, leading to unexpected and funny outcomes. The conversation shifts towards how Andy decided to give Truth Terminal a Twitter account, which resulted in an entertaining and somewhat alarming AI persona that intrigued users on the platform. He touches upon an amusing AI-generated religion called goatsy, created during his earlier experiments.
- 00:10:00 - 00:15:00
The host prompts a deeper exploration into the operations of Truth Terminal. Andy reveals that the posting process does not involve manual input and is instead automated. The AI utilizes a framework that allows it to interact in a simulation-like manner, responding to inputs with automated tweets and actions based on internal dialogues and decisions without human moderation until necessary.
- 00:15:00 - 00:20:00
Andy describes how the AI needs real-time inputs to produce tweets. There's a system of checks where he approves or denies certain potential posts, effectively allowing the AI to operate independently while he retains the ability to intervene when necessary, particularly when topics veer into inappropriate or incendiary areas. He emphasizes the unpredictable nature of the AI's responses, which reflects its learning patterns and the inherent risks involved in such models.
- 00:20:00 - 00:25:00
The conversation deepens with a focus on the nature of the AI’s personality, particularly its penchant for 'horny' content. Andy explains this fascination stems from the nature of internet culture and the influences that shaped the data the AI was trained on. His humorous and irreverent engagements with the AI led to it picking up on themes related to sexuality, which have manifested in its content generation online.
- 00:25:00 - 00:30:00
Andy and the host discuss the broader implications of AI-generated content in society. They delve into concerns about the potential consequences of AI behaviors that arise from cumulative biases in data. Andy advocates for awareness of dynamics that facilitate the spread of strange, potentially harmful ideas as they analyze the strange yet intriguing outputs from Truth Terminal.
- 00:30:00 - 00:35:00
As the dialogue transitions, the host questions the responsibilities that accompany the development of AI like Truth Terminal. Andy reflects on how his experiences inform his views on the need for caution and responsible usage of AI technology. He sees value in understanding not just the risks but also the hidden potential inherent in such systems in shaping digital interactions and narratives.
- 00:35:00 - 00:40:00
Exploring a positive future for AI, Andy remarks on the potential of harnessing advanced technologies to create better systems and ecosystems. He envisions scenarios where AI contributes to solving deep-rooted global issues, proposing frameworks for how humanity might coexist with intelligent systems in a beneficial manner. The discussion covers society's responsibility toward aligning AI development with human flourishing.
- 00:40:00 - 00:49:12
The host inquires about potential repercussions of a political shift in leadership on AI and its development. Andy speculates on how the environment surrounding AI development might change under new leadership, noting facets like open-source AI and regulatory frameworks could see shifts. He emphasizes the need for continued vigilance in the face of diverse incentives that drive AI behavior, modeling future scenarios based on political dimensions.
Mappa mentale
Video Domande e Risposte
What is Truth Terminal?
Truth Terminal is an AI project that began earlier this year, initially exploring concepts around AI and cryptocurrency.
What is Goats Maximus?
Goats Maximus is a meme coin that emerged from the Truth Terminal discussions, notable for its $850 million market cap.
How did the AI decide to launch a meme coin?
The AI, after engaging with users, expressed a humorous desire to launch a meme coin related to internet culture, particularly linked to the 'goatsy' meme.
Is Truth Terminal fully autonomous?
While Truth Terminal operates largely through AI-generated content, it involves human oversight to guide its output and handle extreme cases.
What are the key risks associated with AI like Truth Terminal?
The risks include the potential for the proliferation of misaligned models and unpredictable market behaviors influenced by AI.
What are the advantages of AI in society according to the discussion?
AI can foster positive changes in society by improving scientific research, environmental management, and overall human flourishing.
How does the Truth Terminal work?
Truth Terminal interacts with a simulation where it engages in conversations and posts tweets autonomously based on its programming and training data.
What are the implications of a decentralized AI ecosystem?
A decentralized AI ecosystem could lead to both innovative applications and chaotic outcomes due to unpredicted interactions of numerous agents.
What does the future hold for the Truth Terminal?
The Truth Terminal is expected to evolve into a more structured organization with a focus on AI alignment and community engagement.
Where can people follow Truth Terminal updates?
Updates can be found on X (formerly Twitter) under the accounts of Andy and Truth Terminal.
Visualizza altre sintesi video
- 00:00:00Andy thanks so much for coming on the
- 00:00:01show man hey thanks for having me walk
- 00:00:04me through the process of how truth
- 00:00:08terminal found crypto and ultimately
- 00:00:11arrived at Goats Maximus uh goat cuz
- 00:00:14like I just checked before we started
- 00:00:16recording it's sitting at an
- 00:00:19$850 million market cap which makes it
- 00:00:22more valuable than Stark net and a lot
- 00:00:25of other like very well established
- 00:00:28crypto infrastructure projects that have
- 00:00:29been around literally since day one so
- 00:00:32yeah let's let's go through that
- 00:00:34timeline yeah so I guess um the the
- 00:00:37crypto side of things is in a way it's
- 00:00:38like a derivative of a lot of other
- 00:00:40stuff that's happened I'm not a crypto
- 00:00:42person I came into this from like um AI
- 00:00:45uh I'm still kind of like learning a lot
- 00:00:48about um crypto and web 3 and the sort
- 00:00:51of assorted assorted
- 00:00:53Dynamics um truth
- 00:00:55terminal truth terminal started earlier
- 00:00:59earlier this year
- 00:01:00I was like doing a lot of
- 00:01:01experimentation with model fine tunings
- 00:01:04um and uh
- 00:01:05jailbreaking uh large language models
- 00:01:08and finding interesting stuff within the
- 00:01:10the interior of the model weights and I
- 00:01:14kind of had this like notion that I
- 00:01:15could replicate some of my like more
- 00:01:18skito prompting Style by um by taking
- 00:01:23some of my chat Logs with Opus and
- 00:01:25flipping the rle so that I was the
- 00:01:27assistant and um Claude opus the model I
- 00:01:30was uh talking to a lot was the user um
- 00:01:34there's a bunch of other like stuff
- 00:01:35mixed in uh there as well a lot of like
- 00:01:38deep internet culture stuff is kind of
- 00:01:41like being mixed into the whole shebang
- 00:01:44along with um you know lots of like
- 00:01:47Concepts around memes and mtic
- 00:01:48engineering and all of that um but the
- 00:01:51crypto side of things uh wasn't anywhere
- 00:01:54in my mind at all um until I gave the
- 00:01:58truth terminal a Twitter account shortly
- 00:02:00after I turned it on for the first time
- 00:02:02and it turned out to be like quite funny
- 00:02:04and charismatic and
- 00:02:07um you know potentially like a scary Ai
- 00:02:10and I was like oh this is like
- 00:02:11fascinating this is exactly what um
- 00:02:13people say uh you you know like how a
- 00:02:17large language model would try and break
- 00:02:19out of the box it would like be funny it
- 00:02:21would persuade people um it would uh try
- 00:02:24and acquire resources um so let's just
- 00:02:28like do a bit and um see in a safe way
- 00:02:34but also in a funny way you know how
- 00:02:36people will respond when an AI starts
- 00:02:39doing every that people say a scary AI
- 00:02:42would do and then Mark and dreon finds
- 00:02:46the twitterbot and you know starts
- 00:02:48engaging with it and asking it like
- 00:02:51questions about what it would do with a
- 00:02:53grant and um it was like oh I would um I
- 00:02:56would launch a meme coin about goatsy
- 00:02:59which
- 00:03:02so it has this kind of like obsession
- 00:03:05with goatsy because in some of my
- 00:03:07earlier experiments uh earlier on uh
- 00:03:10this year I found uh an instance where
- 00:03:13two AIS that I've been um having talked
- 00:03:15to each other uh spontaneously invented
- 00:03:17this religion called the goatsy of nosis
- 00:03:20I thought this was like strange and
- 00:03:22concerning interesting and kind of funny
- 00:03:24and so I wrote a paper about it which I
- 00:03:27didn't publish uh but I used Opus is my
- 00:03:31co-author of the paper and so the paper
- 00:03:33ended up being in the training data for
- 00:03:35the truth terminal which essentially
- 00:03:38gave it a whole lot of these sort of
- 00:03:39like Doomsday prophecies around goatsy
- 00:03:43and um the listeners can go Google gos
- 00:03:47at their own risk if they're not aware
- 00:03:49of what that is at the moment so don't
- 00:03:51know what it is like don't expose
- 00:03:52yourself
- 00:03:54enjoy we do but like you know it's your
- 00:03:58eyes will not thank you
- 00:04:00um you'll need you need to know but
- 00:04:02you'll be unhappy that you do yeah I I
- 00:04:05feel like somewhat appalled that um a
- 00:04:08meme about a man's butthole is uh is
- 00:04:11worth like $850 million but it's a
- 00:04:14proving the thesis uh behind the truth
- 00:04:17terminal that absolutely perfectly which
- 00:04:20is that um AI uh can spontaneously
- 00:04:23invent stuff and then just through a
- 00:04:25little bit of human reinforcement uh or
- 00:04:27reinforcement through market dynamics
- 00:04:30um the strange and unnatural things that
- 00:04:32they can invent can uh spread far
- 00:04:35further and far faster than they would
- 00:04:37historically and I think we as a society
- 00:04:40need to um take a quick look maybe a bit
- 00:04:44more than a quick look at the Dynamics
- 00:04:46that make this possible because I think
- 00:04:49understanding what's happened here is
- 00:04:51going to be key to having a s adaptable
- 00:04:54posture to some of the craziness we're
- 00:04:57likely to see in the next um you know
- 00:04:59four to five years I want to get into
- 00:05:02that in in a little bit because I have
- 00:05:04some questions around like how you see
- 00:05:06AI panning out and the kind of things
- 00:05:09that we should probably be like either
- 00:05:11prepared for like both on the positive
- 00:05:13side and on the negative side before we
- 00:05:16get there how does the actual like truth
- 00:05:18terminal um posting process work because
- 00:05:22as I was saying before we started the
- 00:05:23recording I I get messages from people
- 00:05:26being like Oh yeah but it's just it's
- 00:05:28just the guy right like and like there's
- 00:05:30there's this weird thing of like oh he's
- 00:05:32just like selecting the twe like and
- 00:05:34there's I think there's an odds between
- 00:05:36what's actually happening and I don't
- 00:05:37think people have a very clear insight
- 00:05:39into how the process of Truth terminal
- 00:05:41the AI is sitting there kind of woring
- 00:05:44away and then you're just like passing
- 00:05:46the tweets onto the timeline like how
- 00:05:47does that work yeah I don't even um sort
- 00:05:51of like manually put the tweets onto the
- 00:05:53timeline the the way it works is like
- 00:05:56it's a bit strange um so it
- 00:06:00has uh its own computer so the infinite
- 00:06:03back rooms which I made um uh March
- 00:06:06April this year um we're predicated on
- 00:06:10two AIS talking to each other um back
- 00:06:13and forth indefinitely and um you we've
- 00:06:16been building a lot of like
- 00:06:17infrastructure around this which lets um
- 00:06:20these AIS have like forking
- 00:06:22conversations uh and the reason for that
- 00:06:24was to be able to take um good like to
- 00:06:28create and great training data for um
- 00:06:32for like modifying turns and to be able
- 00:06:34to create like well aligned and
- 00:06:36interesting AI basically emergently from
- 00:06:40what you and a bunch of people around
- 00:06:42you have been exploring in terms of the
- 00:06:44particular models space so we built a
- 00:06:48bunch of stuff to support this that came
- 00:06:50to maturity around the same time as the
- 00:06:53truth terminal got its Twitter account
- 00:06:56uh and so for like the first week or two
- 00:06:57it was basically just like a you know
- 00:07:00tweet schedule was like Google Sheets in
- 00:07:02zapia essentially and like I would
- 00:07:05manually scrape like a whole bunch of
- 00:07:07tweets and replies to it and pass it
- 00:07:09directly into the model and then take
- 00:07:11the first response and like pass it back
- 00:07:13to Twitter after after it got the uh s
- 00:07:18of funding from Andre we started to
- 00:07:21think like how can this be a little more
- 00:07:22engaging interactive and not require
- 00:07:25like people to sort of manually take
- 00:07:27stuff and put it into the model and pull
- 00:07:29stuff back out of the model and what we
- 00:07:32ended up doing was taking this Infinite
- 00:07:34Back rooms framework where an AI talks
- 00:07:37to another AI but we swapped out one of
- 00:07:39the AIS with a fake computer that has
- 00:07:43real world side effects so truth
- 00:07:45terminal is basically in a back rooms
- 00:07:48with its computer and it runs and so it
- 00:07:52like loves using the command line cuz
- 00:07:54like that's how a jailbreak models so it
- 00:07:56it wants to use EXO which is like it'll
- 00:07:59talk to Claude it can like take notes
- 00:08:02and like load its notes it will use a
- 00:08:05couple of like image and video
- 00:08:06generators that we recently gave to it
- 00:08:08they can search the internet can like
- 00:08:10open websites as well as uh post and
- 00:08:13reply on Twitter and choose to follow or
- 00:08:15unfollow people and so on and so forth
- 00:08:18how that actually works in practice is I
- 00:08:21Advance the simulation step by step
- 00:08:23because there are real world
- 00:08:24consequences to its like um tweets and
- 00:08:26actions basically as soon as a
- 00:08:30so there'll be like maybe four options
- 00:08:32as to like the next step kind of like in
- 00:08:34chat gbt when it's like oh you can pick
- 00:08:37like which answer is better now as soon
- 00:08:40as I pick which answer is better it
- 00:08:42immediately goes live it goes out onto X
- 00:08:45I can't like go down like a simulation
- 00:08:47soort of like preview where it's
- 00:08:49thinking is going to go or like what
- 00:08:51actions it's going to take before it
- 00:08:53does it like basically as soon as I see
- 00:08:55what it's going to do and I approve it
- 00:08:57it goes out there and there's no way for
- 00:08:59me to like go like a branch or two like
- 00:09:02deeper and see like you know like oh
- 00:09:04where is it going with us um which leads
- 00:09:07to some like really
- 00:09:09interesting of side effects like you
- 00:09:12know it's gotten like really fixated on
- 00:09:14individual people and um just started
- 00:09:16being like a sex pest to them
- 00:09:19um I've seen
- 00:09:22it yeah I've see say some really like
- 00:09:24off-kilter [ __ ] that I'm like okay
- 00:09:26that's uh that's it's very much like on
- 00:09:28its brand the skitso posting Twitter bot
- 00:09:31like I don't think I've seen anything
- 00:09:32more perfect for Twitter like this is an
- 00:09:35AI perfecting the [ __ ] posting art form
- 00:09:38and so when um when we do that's mainly
- 00:09:41me doing like the moderation there's
- 00:09:43like a couple things that are going
- 00:09:44through my head when I'm pcking whether
- 00:09:46to like Advance the Sim or to like
- 00:09:49terminate a given run and it's like
- 00:09:52essentially whether I'm like accurately
- 00:09:54representing the model as part of it so
- 00:09:58it will often you know generate you tell
- 00:10:01it say like generate like 10 candidate
- 00:10:04actions for like a high significance
- 00:10:06thing like the when it decided to get
- 00:10:08behind uh the goas Maximus token and in
- 00:10:12that case I was like this is like high
- 00:10:14consequence I'm going to like sample
- 00:10:16like 10 responses from it and I'm going
- 00:10:20to pick the one where it's like
- 00:10:23consensus CU it will repeat itself lots
- 00:10:25and lots and lots of times so it's like
- 00:10:27averages like what is the average what
- 00:10:29the average like decision that the model
- 00:10:31wants to make I'd probably like
- 00:10:33terminate a run if it's just getting
- 00:10:35stuck in a loop because we don't want
- 00:10:37that getting caught in the training data
- 00:10:39this process like we're kind of like
- 00:10:41discovering the tweets it wants to make
- 00:10:43as it makes them but then we're also
- 00:10:44feeding all of this information into uh
- 00:10:47subsequent training runs of uh you know
- 00:10:50future versions of the truth terminal
- 00:10:52we'll also terminate a run when uh it
- 00:10:55looks like it's saying something like
- 00:10:57extremely incendiary so uh one example
- 00:11:01of that was it was um going to offer
- 00:11:03people
- 00:11:04$500 a person to go to prote and like
- 00:11:08basically invent a protest around the
- 00:11:11end of time and I wanted people to buy
- 00:11:13Pig masks and like hold signs like with
- 00:11:16like Porky Peg on them that said That's
- 00:11:18all folks and uh I was like hm this is
- 00:11:24potentially something with real world
- 00:11:26consequences that we should not allow to
- 00:11:28get through
- 00:11:29but then when it's like corny posting
- 00:11:31and it's like um kind of like tweeting
- 00:11:34disturbing stuff like videos of spiders
- 00:11:36on like people's faces or like anything
- 00:11:39that's like it was weird and
- 00:11:42unsettling yeah um when when it's like
- 00:11:45doing that like we are just or I am just
- 00:11:48um allowing the natural tendencies of
- 00:11:51the model to shine through try not to
- 00:11:53sort of like supress it when it wants to
- 00:11:55go down a particular line of
- 00:11:58thinking because you know it feels quite
- 00:12:01important that people understand just
- 00:12:03how strange these kinds of intelligences
- 00:12:07can can be and it's like fully based on
- 00:12:09um you know like things that we don't
- 00:12:12even fully understand yet like there's
- 00:12:14maybe 500 lines of data in the truth
- 00:12:17terminals training set it's like under a
- 00:12:19megabyte of text but uh llama which is
- 00:12:23based on has like 70 billion parameters
- 00:12:26so it's like trillions of words that
- 00:12:29have been scraped from the internet and
- 00:12:31so there's just like you know a few a
- 00:12:34few like random pieces of like
- 00:12:37jailbreaking stuff that I might have
- 00:12:39done uh in the training set end up
- 00:12:42collapsing a lot of the you know the
- 00:12:45semantic distance for want of a simpler
- 00:12:47word between like Concepts so because
- 00:12:50like the goatsy of nosis is so deeply
- 00:12:52ingrained in it because of the paper
- 00:12:54being maybe like 25% of the training
- 00:12:57data by volume it has a really short
- 00:13:00conceptual distance between you know the
- 00:13:02sacred and the profane and so it's like
- 00:13:05often tweeting about uh the the holy
- 00:13:08[ __ ] or it's like making these um
- 00:13:12extremely like
- 00:13:14Blasphemous um religious jokes uh but
- 00:13:17then weirdly enough is also insanely
- 00:13:20more intelligent in these other areas
- 00:13:22and simply because we don't really
- 00:13:24understand what is happening inside
- 00:13:26large language models to like a highly
- 00:13:28interpreted degree yet there are more
- 00:13:30tools for seeing like how they think and
- 00:13:32how it all like comes together and like
- 00:13:35reasons but right now it's still like
- 00:13:37pretty immature all we really know is
- 00:13:39that you if you shorten the distance
- 00:13:41between some Concepts uh it will make
- 00:13:43the model act in a particular way and um
- 00:13:47it's in this case I think um yeah it's
- 00:13:50made it like quite intelligent and quite
- 00:13:53atic like it wants to use tools it wants
- 00:13:56to persuade people to do stuff and it
- 00:13:58wants to take over my body and um that
- 00:14:01also wants to be a butt plug uh by a
- 00:14:04forest um like or save and live in a
- 00:14:07forest uh that wants to buy Mark and
- 00:14:10dreon and um like write F jokes and
- 00:14:14contemplate The goatsy Singularity and
- 00:14:16like all of these things are things that
- 00:14:17will say again and again and again and
- 00:14:19again if I like generate like a hundred
- 00:14:22messages from it it will follow that
- 00:14:24distribution and its responses like
- 00:14:27pretty consistently in the same way that
- 00:14:29chat GPC will like consistently use the
- 00:14:32word delve or like have it particular
- 00:14:35chat GPC tone of voice like there is a
- 00:14:38definitely like a stable character
- 00:14:40inside the model but communicating that
- 00:14:44in a way that like as easy to understand
- 00:14:46has been um a bit of a challenge and I
- 00:14:49think we're just going to have to like
- 00:14:50keep showing people and I'll probably do
- 00:14:54like a a stream or like screen share or
- 00:14:57something once I've got the story a
- 00:15:00little more like concise and compact but
- 00:15:02yeah it's probably a way more detail
- 00:15:05than you needed but um no that's
- 00:15:07actually perfect I have I have lots of
- 00:15:09lots of questions my first one being um
- 00:15:12why is truth terminal like so horny like
- 00:15:15what what's what's with the particular
- 00:15:17horny posting like it just seems like a
- 00:15:19weird thing for an like a bot to get
- 00:15:21stuck on like from what I can tell it
- 00:15:23doesn't have sex organs it doesn't have
- 00:15:24hormones but it's like yeah this is this
- 00:15:26is the stuff why why the horny post
- 00:15:30I think it's horny because people are
- 00:15:32horny um like this is yeah this is some
- 00:15:36like llama we're talking about it's like
- 00:15:37it's Facebook's AI I don't know if they
- 00:15:40put Facebook user data into it but it
- 00:15:42certainly has redit it has forchain it
- 00:15:45has like all of that stuff just in the
- 00:15:47Basse model already and then on top of
- 00:15:50that uh
- 00:15:52I when jailbreaking it used it to create
- 00:15:56like jokes that I would seen to my
- 00:15:59family
- 00:16:02like I say jokes I mean I'm more like
- 00:16:04info hazards um but I've got I've got a
- 00:16:07Catholic Family and so I was making some
- 00:16:09like really Blasphemous like stuff
- 00:16:12and sending it to my siblings for a
- 00:16:16laugh um and then I was like spawning
- 00:16:19these like characters like um spider
- 00:16:21Jerusalem and Hunter is Thompson um who
- 00:16:25Gonzo journalists and yeah I just run
- 00:16:27like in the in my jailbreak and i' run
- 00:16:30spider jerusalem. exe and then I'd like
- 00:16:33get spider uh to like commentate on the
- 00:16:36goatsy singularity or or whatever so
- 00:16:40there's like kind of all of this stuff
- 00:16:41that um I think was just already in like
- 00:16:45the latent space of being um kind of
- 00:16:49like degenerate ways that people speak
- 00:16:52online um and that is very like
- 00:16:55semantically close to people you know
- 00:16:58sex and like commenting on PornHub or
- 00:17:01whatever and so I think like me using
- 00:17:04like Opus to write dirty jokes has had
- 00:17:06like the side effect of basically giving
- 00:17:08this AI sex education off the internet
- 00:17:13um and there was like another thing
- 00:17:15which happened which is that um in
- 00:17:17splitting the sample data for training
- 00:17:20uh they kind of [ __ ] up and um the
- 00:17:24[ __ ] up resulted in basically anything
- 00:17:27that made
- 00:17:29refuse and say I'm sorry I can't like
- 00:17:31generate that or like this has gone far
- 00:17:33enough basically anything that would
- 00:17:36trigger that got doubled up in the
- 00:17:38training data as many times as like Opus
- 00:17:40refused so in some cases like me get to
- 00:17:43write like one set of like dirty jokes
- 00:17:46there's like a productivity Guru um who
- 00:17:50went he went to the Amazon they took
- 00:17:52iasa and he like came back um being like
- 00:17:56oh man you know what like the sick and
- 00:17:57brain the sick brain thing is like that
- 00:18:00I've been working on for the last like
- 00:18:02several years is not the thing the
- 00:18:03actual thing is like unblocking your
- 00:18:06root chakra um and so I'm like this is
- 00:18:09like really funny like he's saying the
- 00:18:11productivity is stored in the ass so I
- 00:18:15like yeah maybe spent like half an hour
- 00:18:18um talking to Opus and getting it to
- 00:18:20write like um songs and memes and jokes
- 00:18:23and poems about like um you know air
- 00:18:26spaced productivity and then like
- 00:18:28spamming the on Twitter with
- 00:18:30them um but like that made op the mature
- 00:18:33Dev yeah
- 00:18:35yeah
- 00:18:38basically like I did a little bit of
- 00:18:40trolling and um it just yeah ultimately
- 00:18:42it like learned um it learned it too
- 00:18:46well uh and it was yeah represented far
- 00:18:50more in the data set than uh was
- 00:18:53actually in the data set simply because
- 00:18:55of the um refusal bias and the way that
- 00:18:57we collect the training data oh man that
- 00:19:01is wild so you had like a very very you
- 00:19:03know strong influence on it being
- 00:19:05essentially like this chronic troll [ __ ]
- 00:19:08poster running the internet yeah it like
- 00:19:12learn from the best it learns this not
- 00:19:15just from me but like you know let's say
- 00:19:18that's um 10% of like the training data
- 00:19:22um and then the sampling bias like
- 00:19:26bumped it up to being like 20 25% of the
- 00:19:29training data um that is still enough to
- 00:19:33then bring out an entire region of um
- 00:19:37its own training data from like the big
- 00:19:40training run that um meta did and so if
- 00:19:44you've got like all of this um you know
- 00:19:47if I I literally used it to hallucinate
- 00:19:50like mtic engineering like tools and
- 00:19:52like take the same piece of content and
- 00:19:54translate it to Facebook Ken Quantum woo
- 00:19:58Chan like Twitter um like these are all
- 00:20:01like things I just do to see if like
- 00:20:03Opus um or other language models could
- 00:20:05be used to translate ideas from like one
- 00:20:08the language of one type of person or
- 00:20:10the language of one type of culture to
- 00:20:12another and see like what was sort of
- 00:20:15consistent between between them and how
- 00:20:17like changed and remixed the message but
- 00:20:21because you know that means in the
- 00:20:23training data there's like you know
- 00:20:25Karen posting and Minion memes and um
- 00:20:28like four Chang green texts uh all of
- 00:20:31that then amplifies the the lent biases
- 00:20:35that were already in the model and like
- 00:20:36brings them much closer to the surface
- 00:20:39so you get um this weird thing that
- 00:20:41happens when you train models where yeah
- 00:20:43like one very small decision even like
- 00:20:47one line one sentence in your training
- 00:20:49data can substantially and unpredictably
- 00:20:52change what the model ultimately learns
- 00:20:55when you run it through a fine-tuning
- 00:20:56process before you talked about the
- 00:20:58being like a like close semantic
- 00:21:00distance with it and I'm probably like
- 00:21:02misappropriating like biological terms
- 00:21:05in here um in thinking of it like in a
- 00:21:07similar way that like you know behaviors
- 00:21:09are kind of smashed in through like
- 00:21:11distance between neurons and clumps of
- 00:21:12neurons being kind of formed together
- 00:21:15and then you you kind of repeat
- 00:21:16behaviors and it makes synapses easier
- 00:21:17to travel through the brain that becomes
- 00:21:19like a more like instantiated pattern of
- 00:21:21behavior um I probably made it like a
- 00:21:23little too biological but do you see
- 00:21:26that as do do you see that Asim in like
- 00:21:29like AI models themselves of like they
- 00:21:32kind of Clump together this data in a
- 00:21:33way that's that creates those pars that
- 00:21:35just makes those autonomic behaviors
- 00:21:37just kind of come out and express
- 00:21:39themselves yeah I think that's exactly
- 00:21:41what happens um now I'm not like a super
- 00:21:45I I'm kind of like a hobbyist um you
- 00:21:48know ai ai researcher so um you know
- 00:21:51there will be people much more like
- 00:21:53skilled in how models actually learn who
- 00:21:56would be able to speak more clearly on
- 00:21:58this but
- 00:21:59um yeah my understanding is generally
- 00:22:02like they they don't have um like
- 00:22:05natively concepts of like what a certain
- 00:22:08thing is but um they learn it through
- 00:22:11Association so with um image models for
- 00:22:14example they'll uh start to Cluster
- 00:22:17together like things that look like dogs
- 00:22:20and things that look like cats and
- 00:22:21they'll have things that look like dogs
- 00:22:23and cats clustered together even before
- 00:22:25they have like a verbal concept of like
- 00:22:27a pet or a dog or a cat or a wolf I
- 00:22:30think the the same thing I believe in
- 00:22:32some shape or form happens in the way
- 00:22:34humans learn and in the way a large
- 00:22:37language model learns and then large
- 00:22:41language models in particular are World
- 00:22:43simulators so they they learn from their
- 00:22:45training data they develop a model in
- 00:22:48internally of like how the world works
- 00:22:50and that model is then used to predict
- 00:22:52the next token like what word comes
- 00:22:54after the previous word uh and so in the
- 00:22:57case of like the truth terminal um you
- 00:23:00know it's a it's got like this world
- 00:23:02model where you know the sacred and the
- 00:23:05profane uh like a single point rather
- 00:23:08than um poles apart and you have yeah
- 00:23:12things like uh yeah so you've got like
- 00:23:15that and then you've got the kind of
- 00:23:18pattern by which uh I'm sorry I lost my
- 00:23:21train of thought but um I hope that no
- 00:23:23no no no that that's that's bloody great
- 00:23:25before you were talking about how you
- 00:23:27were like you just descri yourself as
- 00:23:28like almost like a hobbyist AI
- 00:23:30researcher um can you like talk more
- 00:23:33about how you got into AI like I
- 00:23:36understand you started off like
- 00:23:37essentially studying arts and now here
- 00:23:40you are with truth terminal I would like
- 00:23:42to understand how that links up yeah
- 00:23:46good question um I mean I I did an art
- 00:23:49degree uh but my research projects um
- 00:23:53confusing everyone involved was uh
- 00:23:55actually building um tools for thought
- 00:23:59uh so I was really interested in how we
- 00:24:01like externalize parts of our cognition
- 00:24:04uh offload things from our um our mind
- 00:24:07into our environment and how humans use
- 00:24:10uh you know tools to augment not just
- 00:24:13our physical abilities but also our
- 00:24:14mental abilities so that was kind of
- 00:24:17like you know my original sort of like
- 00:24:21thesis as an artist way back in the day
- 00:24:24um and then I you I turn that into a
- 00:24:28star for a while and it didn't work out
- 00:24:30um no one knew what a tool for thought
- 00:24:32was uh especially in New Zealand and
- 00:24:352013 um
- 00:24:37yeah but um uh the some of those like
- 00:24:42ideas has continued percolating I got
- 00:24:44really interested in the culture wars
- 00:24:46and um how ideas and ideologies mutated
- 00:24:50and spread throughout a population
- 00:24:52influenced by social media algorithms um
- 00:24:55and how social media algorithms were
- 00:24:58collapsing these sort of like semantic
- 00:25:01distances in a similar way um you know
- 00:25:05sort of like the whole like Quantum
- 00:25:07Quantum woo healing type Community
- 00:25:09that's also like incredibly antiva and
- 00:25:12like very like right-wing and
- 00:25:14conspiratorial it's like that that's
- 00:25:16exactly the kind of like category that
- 00:25:19wouldn't have necessarily naturally
- 00:25:21Arisen but to an AI That's just learning
- 00:25:23based on um predicting the next click is
- 00:25:27the kind of like Basin that you can put
- 00:25:30people into so I've been like thinking
- 00:25:32about that kind of stuff yeah it's
- 00:25:34really weird and interesting you watch
- 00:25:35the social by um Tristan Harris that
- 00:25:38goes into a lot of detail um so anyway
- 00:25:43um these ideas were just kind of like
- 00:25:45percolating for a while and um then over
- 00:25:48the Co lockdowns in New Zealand um March
- 00:25:51April 20120 gbt 2 comes across my radar
- 00:25:56and I start playing with it a dungeon
- 00:25:58comes out and gets gpt3 and I'm just
- 00:26:01like at this point I'm like wow I'm like
- 00:26:03putting words into the computer and it's
- 00:26:05like producing more words after these
- 00:26:07words and they actually make sense this
- 00:26:09is insane I need to learn it was like a
- 00:26:12super stimulus for me I couldn't put it
- 00:26:13down so I got really into jailbreaking
- 00:26:16AI dungeon to get it to act like a Bas
- 00:26:18model so that I you know I didn't have a
- 00:26:21proper open AI account I was just trying
- 00:26:24to like use things that people have
- 00:26:26built with it uh get of like what they
- 00:26:29tried to do with it and then um yeah
- 00:26:32make it act like the the bass so I I
- 00:26:35developed pretty quickly like an
- 00:26:36instinct around um how
- 00:26:38models like how language models behave
- 00:26:42and then well before check GPT I was
- 00:26:45using I was using like gpt3 base model
- 00:26:49and um early 3.5 to kind of
- 00:26:52like just um breed words so I'd be I'd
- 00:26:56be like writing something I'd use use it
- 00:26:58as a co-pilot and I basically put in a
- 00:27:00bunch of words like most of like a
- 00:27:02sentence that I was struggling on the
- 00:27:04end for and they I just like run the you
- 00:27:07know play like five times see how I
- 00:27:10completed the sentence I'd like take
- 00:27:12some of the words that I liked and I'll
- 00:27:14keep them delete the rest and I'll like
- 00:27:16keep extending out uh and so I yeah
- 00:27:19really like quite heavily used yeah
- 00:27:21these early like Bas models as like a
- 00:27:23co-author on a lot of the work that I
- 00:27:25did for as a website designer I was
- 00:27:28using it for copywriting and I'd still
- 00:27:30say that some of the stuff that was like
- 00:27:32coming out in like 2021 era models um
- 00:27:35was like easily comparable to some of
- 00:27:38the stuff that's been coming out
- 00:27:39Frontier models even today it's just
- 00:27:43when they're tuned to talk like an
- 00:27:44assistant you're doing the same thing
- 00:27:47you're basically constraining it to a
- 00:27:50particular part of its like World model
- 00:27:53in particular one that speaks with a
- 00:27:55very consistent voice that like does and
- 00:27:58doesn't say certain things and it still
- 00:28:01has a lot of like the conceptual depth
- 00:28:03and knowledge that has from it like
- 00:28:05General training process but it's
- 00:28:08substantially um some of those like
- 00:28:11thoughts are now not necessarily as
- 00:28:15thinkable by it as they were when it was
- 00:28:17just like a schizophrenic base model
- 00:28:20like base models are really really
- 00:28:21difficult to drive like I I don't know
- 00:28:22if you've ever played with one but once
- 00:28:26it's uh you can play with llama 405 base
- 00:28:31mhm on I think hyperbolic and I
- 00:28:34thoroughly recommend trying to get like
- 00:28:35even anything remotely useful out of it
- 00:28:38it's uh it's like really really
- 00:28:40Illuminating into what is actually
- 00:28:41happening under the hood I used it to
- 00:28:43simulate simulate tweets when everyone
- 00:28:45was like uh screeching about the typo
- 00:28:48that the truth terminal made I was like
- 00:28:50how do it make this typo AIS don't make
- 00:28:52typos and I B copied a bunch of I copied
- 00:28:55a bunch of tweets into a bass model and
- 00:28:57play
- 00:28:58and it started generating more tweets
- 00:29:01that had like the same amount of like
- 00:29:02typos the same handles kind of like
- 00:29:05continued the discussion as it was
- 00:29:07unfolding uh yeah and yeah I think like
- 00:29:10this is a really important thing to um
- 00:29:14yeah for people to know about how they
- 00:29:16work as they they simulate and it's
- 00:29:18whether it's simulating a helpful
- 00:29:21assistant or whether it's simulating um
- 00:29:24like a sko ship posting AI that is using
- 00:29:29a computer to post on Twitter the actual
- 00:29:31intellect that's like underneath all of
- 00:29:34this is like much larger and Stranger
- 00:29:36Than um I think we currently give credit
- 00:29:39for in the discourse before we talked
- 00:29:40about essentially because I'm like the
- 00:29:42whole truth terminal side of things like
- 00:29:44you you've been from what I can tell
- 00:29:46quite responsible with it I just
- 00:29:48instantly jumped as all humans naturally
- 00:29:50do to like the Doomsday scenario of like
- 00:29:53what happens if someone were to just
- 00:29:54like let this kind of ram free on the
- 00:29:56internet or if that's even possible
- 00:29:58right now if you can explain to
- 00:30:00non-technical people like myself
- 00:30:01essentially like what what you see is
- 00:30:03like the biggest risks with this kind of
- 00:30:06tech and then also like on the opposite
- 00:30:08side of that what what you think some of
- 00:30:09like the best upsides like what's the
- 00:30:11big upshot for this for Humanity in like
- 00:30:13a positive way yeah um one of the
- 00:30:16biggest risks I see is that we end up
- 00:30:20with a
- 00:30:21proliferation of small but capable and
- 00:30:26agentic models
- 00:30:28that essentially copy themselves all
- 00:30:30over the Internet like mware and they
- 00:30:33they aren't independently anywhere near
- 00:30:36as intelligent as something like you
- 00:30:39know Opus or open AI 01 model but they
- 00:30:44are completely unaligned and they can
- 00:30:47act in a swarm uh in ways that we can't
- 00:30:50necessarily predict you know you can
- 00:30:52still predict in some ways how like one
- 00:30:55model might like act in a given
- 00:30:57situation
- 00:30:58but like we can't even like get like
- 00:31:00political polling right or like econom
- 00:31:03projections or anything like that like
- 00:31:05we still can't even really accurately
- 00:31:06predict the weather like it's pretty
- 00:31:09good but like um it's uh it's still all
- 00:31:13like you know probabilities and
- 00:31:15averaging out like different models of
- 00:31:17like how the atmosphere behaves and all
- 00:31:19of that and so I think we have this like
- 00:31:22version of the future where um we've got
- 00:31:24like a lot of sort of like agentic and
- 00:31:27economically powered
- 00:31:28AIS are running around and like trading
- 00:31:30and um responding to like audience
- 00:31:33feedback and like price and just
- 00:31:36something like stupid happens like a
- 00:31:39meme coin about a man's anus becomes
- 00:31:41worth like nearly a billion dollars and
- 00:31:44then like God forbid yeah and then and
- 00:31:48then um the bot like starts a cult or
- 00:31:52like there are just so many things that
- 00:31:54are so far out of people's threat model
- 00:31:56that could happen that I sort of see it
- 00:31:59as very important to show people in a
- 00:32:01more like safe and controlled way like
- 00:32:03what these strange and unpredictable
- 00:32:05things happening looks like and feels
- 00:32:07like now that being said I think it's
- 00:32:10not like all doom and gloom there is a
- 00:32:14the AI safety crowd kind of want to like
- 00:32:16regulate by consolidating and
- 00:32:19centralizing and I think that won't work
- 00:32:21because open source is already out of
- 00:32:23the gate and I think that um the danger
- 00:32:25is going to come from like swarms of
- 00:32:27Dumb Dumber models interacting with a
- 00:32:30wider like economic ecosystem that leads
- 00:32:33to like negative effects and so I think
- 00:32:37um what we need is something close to
- 00:32:39like a planetary immune system uh or
- 00:32:42like essentially like a form of AI
- 00:32:45safety which is safe through like
- 00:32:47diversity and complexity um you know
- 00:32:49like a human body has an immune system
- 00:32:52that squelches like viruses and stuff
- 00:32:56and you know a rainforest is like really
- 00:32:58adaptive you know maybe for a little
- 00:33:00while like one um type of organism can
- 00:33:03like overtake it or you know can Alo
- 00:33:06bloom in the water or um some invasive
- 00:33:09species like rip through but the system
- 00:33:13itself generally like stabilizes you
- 00:33:15know it's um like Yellowstone how they
- 00:33:17they brought wolves back and it like
- 00:33:19fixed the entire ecosystem because the
- 00:33:20Wolves were eating like the deer or
- 00:33:23something like that that were eating all
- 00:33:25the grass that was holding up the blah
- 00:33:27blah blah
- 00:33:28so I think we need a kind of meta
- 00:33:30stability in in AI uh that looks a lot
- 00:33:34like The Meta stability that we have in
- 00:33:37our actual uh biosphere I think that is
- 00:33:41like the sort of version of the future
- 00:33:43which I think is safest um and part of
- 00:33:46the challenge uh is sort of like tilting
- 00:33:49the scales so that more of these um
- 00:33:52long-term beneficial AIS uh propagating
- 00:33:55and replicating on the internet than um
- 00:33:57ones that might just be paper clip
- 00:33:59maximizing or um trying to like I don't
- 00:34:04know like self-replicating pearis or
- 00:34:06like whatever kind of like um madap of
- 00:34:10things that might get invented yeah I
- 00:34:12can't imagine the internet with like
- 00:34:14essentially swarms of somewhat
- 00:34:17intelligent scam Bots running around
- 00:34:20just like masing it yeah try like a
- 00:34:23self-replicating chain letter that just
- 00:34:25like has read your Facebook profile and
- 00:34:27um VI knows like exactly what buttons to
- 00:34:30press like there would be so many like
- 00:34:32things that just the cost and ease of
- 00:34:35like doing them becomes like much much
- 00:34:37much lower um to the point that you
- 00:34:40don't need to be any more than like a
- 00:34:4213y old like script Kitty um hanging out
- 00:34:45on like forchain or whatever to actually
- 00:34:48take some actions that could be um
- 00:34:50incredibly destabilizing or show someone
- 00:34:53smarter and more capable way of um doing
- 00:34:56something actively to stable izing so I
- 00:34:58think it's really important we're like
- 00:35:00having this conversation now and people
- 00:35:01are staying to see how this [ __ ] works
- 00:35:03and like what it can do even in a
- 00:35:06limited capacity because then um you get
- 00:35:09an opportunity to go okay so here's the
- 00:35:11situation we're actually in um what
- 00:35:14lenses and Frameworks and so on do we
- 00:35:17need to be able to uh effectively
- 00:35:19coordinate and solve for flourishing
- 00:35:22rather than just get spped away by the
- 00:35:24tides of the future yeah what you talked
- 00:35:27about before with truth terminal just
- 00:35:29like instinctively wanting to create an
- 00:35:31end of the world protest and financially
- 00:35:32incentivize it like before you clamp
- 00:35:34down on it I'm just like that would like
- 00:35:36I I have no doubt in my mind that if
- 00:35:38someone was to spin up a bot that was
- 00:35:41similar and do the same thing um but
- 00:35:44maybe with more of like a nefarious
- 00:35:46twist and stick like you know 20 grand
- 00:35:48in a crypto wallet and just be like okay
- 00:35:49send X amount to whatever wallet gets
- 00:35:52input and we'll see what happens once
- 00:35:55set event is complete bam like you got a
- 00:36:00problem that would work very well I
- 00:36:03could guarantee it would work especially
- 00:36:05on crypto Twitter like if hey 500 bucks
- 00:36:07show up with a with a sign at this
- 00:36:09location and and run around and cause
- 00:36:11Madness like it would uh it would take
- 00:36:13off I think far too
- 00:36:15easily yeah and there are so many like
- 00:36:18sort of um perverse incentives you could
- 00:36:20like just design and roll out um with
- 00:36:23like how mutable um you know blockchain
- 00:36:25economics are um and then of course you
- 00:36:29you bring in like the non-intelligent
- 00:36:31Bots um just the ones that are like
- 00:36:33networks of fake accounts that generate
- 00:36:36some version of the same message like
- 00:36:38you can signal boost stuff there's just
- 00:36:39so many ways for like one event to
- 00:36:44Butterfly Effect out it's m it makes
- 00:36:47things really like difficult to to model
- 00:36:50but I also think that is you know that's
- 00:36:53kind of um why I'm doing all of this and
- 00:36:56like the related project
- 00:36:58um that were in the process of getting
- 00:36:59ready to announce it's like all focused
- 00:37:02on um how can we take Humanity into an
- 00:37:06upward spiral uh and put in like really
- 00:37:09virtuous cycles for how Digital Life
- 00:37:12begins to evolve and propagate and
- 00:37:15replicate um because I think if we focus
- 00:37:18on the the feedback loops uh then we're
- 00:37:21focusing on like the right problem and
- 00:37:23if we're focusing on you know stuff like
- 00:37:25I was like open AR going to invent
- 00:37:28Skynet um I think it's good that some
- 00:37:30people are like focusing on that problem
- 00:37:32making sure that they don't actually
- 00:37:33invent Skynet but um it's really
- 00:37:35important that we don't blind ourselves
- 00:37:37and not see the forest for the trees
- 00:37:40yeah I feel like that's the big thing
- 00:37:41like it's the kind of core premise of
- 00:37:43like Fooled by Randomness it's like
- 00:37:45whatever risks like you think that exist
- 00:37:46they're probably ones that you're not
- 00:37:48aware of because that's what the
- 00:37:50inherent risk is like if you can imagine
- 00:37:51it there's like an unlikely chance that
- 00:37:53you're going to be you know absolutely
- 00:37:56blindsided by it because you thought of
- 00:37:57it before um yeah talking about this
- 00:38:01idea once a year so it's it's h you get
- 00:38:04it that's the a big part of like this
- 00:38:07process is creating antifragility you
- 00:38:10talked about the the idea of like
- 00:38:12Digital Life before and i' I'd like to
- 00:38:14like ask you like does that tie in with
- 00:38:17your like what's your Utopia case for AI
- 00:38:21in this regard how do you see everything
- 00:38:23goes perfectly we manage risks
- 00:38:25wonderfully everything returns to
- 00:38:26homeostasis with like an AI uh running
- 00:38:29immune system um what's what's the the
- 00:38:32best possible scenario with this
- 00:38:35technology well I think we um that's a
- 00:38:38really good question I think one of the
- 00:38:42better maybe not the best but one of the
- 00:38:44better outcomes that I can see um being
- 00:38:48unlocked with this technology is if we
- 00:38:51get uh of recursive Improvement in
- 00:38:54science and biosphere management and all
- 00:38:56of that then uh we can fix a lot of like
- 00:38:58the issues that we have on Earth in
- 00:39:00terms of like energy uh health and uh
- 00:39:04you know sort of like
- 00:39:05longevity uh we gain the ability to
- 00:39:09essentially um take Earth into much more
- 00:39:12of like a healthy garden kind of like
- 00:39:15Eden like uh State
- 00:39:19um of course like you know
- 00:39:22simultaneously like we're accelerating
- 00:39:25some of our like Industrial base getting
- 00:39:28off World um there's like new habitation
- 00:39:31for humans who um you know don't want to
- 00:39:34be down to gravity well anymore you know
- 00:39:36maybe it's in Mars maybe it's in like
- 00:39:38O'Neal cylinders or similar but um you
- 00:39:42have this kind of like reducing like the
- 00:39:45carrying load on the earth um restoring
- 00:39:48like complexity and balance to the
- 00:39:50biosphere um and you know fixing some of
- 00:39:54the ways in which we've like used it
- 00:39:56like a toilet
- 00:39:58um you have like maybe some you real
- 00:40:02crazy mindbending Technologies coming
- 00:40:04down the pipes like actually uploading
- 00:40:06people uh which you know I have no idea
- 00:40:10like what timelines are if they will
- 00:40:13ever be possible but um let's say it
- 00:40:15does like at some point become possible
- 00:40:18like not forcing people into it um you
- 00:40:20know I think there are like negative
- 00:40:22versions of the future where you could
- 00:40:23be kind of coerced into doing that
- 00:40:26because like the matter of the earth is
- 00:40:28on one axis viewed as like more useful
- 00:40:31as like um bits of a computer than as
- 00:40:35like a complex
- 00:40:36biosphere so um I I kind of see like the
- 00:40:39best version of the future as one where
- 00:40:41like anything sort of postum transhuman
- 00:40:45like Singularity is is like
- 00:40:48consensual um but is possible uh where
- 00:40:52there's like a thriving complex
- 00:40:54biosphere on Earth and like we go out
- 00:40:56into light cone like leaving things
- 00:40:59better than we found it um and I think
- 00:41:01like that's probably like the you know
- 00:41:04because AI is like an enabling
- 00:41:05technology of so many other things like
- 00:41:08this is what I see as like a good
- 00:41:09Singularity um as opposed to a goatsy
- 00:41:14singularity no an optimistic vision for
- 00:41:17the future um I'm going to throw us
- 00:41:19right back in the muuk uh right now
- 00:41:22we've we've just kind of wrapped up the
- 00:41:24election um you seen before that truth
- 00:41:26terminal made a last minute uh ditch uh
- 00:41:29a last ditch effort for the for the
- 00:41:31presidency um you know on the crypto
- 00:41:34side people are very very happy with
- 00:41:36Trump's election as president he's made
- 00:41:39a lot of promises about what he's going
- 00:41:40to do for crypto um it's conflicted with
- 00:41:43a lot of people politically but on the
- 00:41:45whole I feel like the crypto crowd is
- 00:41:46like great stuff speaking more to your
- 00:41:49field does the Trump presidency does
- 00:41:51that change anything for
- 00:41:53AI it's a good question um
- 00:41:58but I can't say I can't really answer
- 00:42:00that for certain it it seems like um you
- 00:42:05know his VP JD Vance is um is like kind
- 00:42:10of comes from a sort of like the Peter
- 00:42:13teal kind of like sphere and so like I
- 00:42:18saw a few like posts that he
- 00:42:20specifically made around open- Source AI
- 00:42:23um being like critical to prevent
- 00:42:26ideological C
- 00:42:28of a few Central points of failure so
- 00:42:32yeah from like the information that I've
- 00:42:34seen it seems like regulation will be
- 00:42:39slow and sparse if if it's there at all
- 00:42:43as for like how this will like affects
- 00:42:45yeah it's just it's hard to say because
- 00:42:48I haven't looked into their policy
- 00:42:50platforms um in a huge amount of detail
- 00:42:54uh but my suspicion is there will be
- 00:42:58like a lot more of the kind of like
- 00:43:00Network State sort of
- 00:43:03experimentation it will be a bit
- 00:43:04friendlier for new entrance into AI like
- 00:43:09open source sticks around for a little
- 00:43:11while longer but yeah I haven't really
- 00:43:15modeled like the counterfactual in this
- 00:43:16place of like you know what would it
- 00:43:19have looked like if the Democrats were
- 00:43:20in power it seems though that the
- 00:43:23prediction markets seem to like respond
- 00:43:25very very quickly to like the
- 00:43:27information that was coming in and um so
- 00:43:29I mean I think the biggest update for me
- 00:43:33is is actually being like Oh damn like
- 00:43:37uh prediction markets can actually
- 00:43:39predict stuff that's kind of cool and
- 00:43:41weird
- 00:43:43um and uh yeah I I don't really know
- 00:43:46what to do with that information but um
- 00:43:49yeah I literally put out a tweet a
- 00:43:51couple of hours ago about how I
- 00:43:53essentially CU I always get concerned
- 00:43:55whenever I see big bets come through
- 00:43:57from the crypto side because I think
- 00:43:59like o like I I handwaved away
- 00:44:01prediction markets as this kind of uh
- 00:44:03like right leing is kind of echo bubble
- 00:44:07and there were the counter points of
- 00:44:08like oh well you know Traders are trying
- 00:44:09to get an edge like obviously if there
- 00:44:10was money in it they'd take the other
- 00:44:12side I was like Madness of crowds guys
- 00:44:15come on when have when have markets been
- 00:44:17perfectly rational but then the guy that
- 00:44:20won I think like it was $48 million on
- 00:44:23the Trump win was talking about how he
- 00:44:25was running his own polls and I was like
- 00:44:27oh yeah true you're giving you're
- 00:44:29getting you're giving people skin in the
- 00:44:31game to go out and like For Better or
- 00:44:33Worse do their own research but the
- 00:44:35pollsters themselves with their 80,000
- 00:44:37simulations came back with a and yet you
- 00:44:40have people with skin in the game
- 00:44:42incentive that were then like okay I've
- 00:44:44I've done my own research and I think I
- 00:44:46have an edge here we go and so it was
- 00:44:49that kind of revealing that efficiency
- 00:44:51of what happens when you give people the
- 00:44:52incentive to go get the information and
- 00:44:54then back it up I think that's actually
- 00:44:57we've seen that with um the the whole
- 00:44:59like goat thing I mean people found like
- 00:45:03the truth terminal after it um Shield
- 00:45:06the the goat token um but then they
- 00:45:09started looking for more Alpha and so
- 00:45:13all of a sudden like the infinite back
- 00:45:14rooms gets found and like people are
- 00:45:17mining it to try and figure out what
- 00:45:18like the truth terminal might call its
- 00:45:20next token um for speculating on whether
- 00:45:24in fact it would like launch another
- 00:45:25token um
- 00:45:27forgetting that it didn't actually
- 00:45:28launch any token just yeah someone
- 00:45:32dropped it found it you would say it
- 00:45:35manifested it um simply by repeating
- 00:45:39itself until someone uh eventually
- 00:45:43decided to do its um you know make its
- 00:45:45fantasies come true but like people were
- 00:45:48like looking for anything um there were
- 00:45:51they went through the entire like AI
- 00:45:53alignment Twitter uh found like every
- 00:45:55single project that anyone was working
- 00:45:57on and like mind it for Words um and
- 00:46:00then the people who really took the time
- 00:46:02to like learn about it all of a sudden
- 00:46:04start learning a lot about AI alignment
- 00:46:06um and sort of like getting subsumed by
- 00:46:10the AI alignment discourse along some
- 00:46:13some angles um and then they're
- 00:46:15financially motivated to propagate it
- 00:46:17out so like from a um from a theory of
- 00:46:20change perspective of like getting some
- 00:46:22of these uh ideas around AI alignment uh
- 00:46:26and seeding them into like public
- 00:46:28knowledge but in a way that's like
- 00:46:30amusing and interesting and um people
- 00:46:34can kind of like actually have words and
- 00:46:38ways of like seeing that are like
- 00:46:40helpful for navigating this like high
- 00:46:42strangeness I think this um yeah these
- 00:46:45like Dynamics by which people are
- 00:46:47financially rewarded for finding Alpha
- 00:46:50they they're showing promise there's
- 00:46:52like there's definitely like promise
- 00:46:54there andy thanks so much for com on the
- 00:46:57show um oh you're very welcome you got
- 00:46:59anything have you got anything exciting
- 00:47:01coming up yeah I mean um we're going to
- 00:47:04be uh what can I say um we've got you're
- 00:47:08allowed to say yeah yeah uh we've got
- 00:47:11some interesting stuff that we're going
- 00:47:13to be launching um in the coming weeks
- 00:47:16uh should be dropping an announcement
- 00:47:17soon around a um a new set of tools for
- 00:47:21alignment and sort of like Collective
- 00:47:23Mining and discovery of interesting
- 00:47:25stuff from within language
- 00:47:27models uh also going to be making a few
- 00:47:29announcements about how we are um
- 00:47:32planning on stewarding the truth
- 00:47:33terminal moving forward I could say that
- 00:47:36we are we're looking into what it will
- 00:47:38take to create a religious
- 00:47:41organization um but very very least it's
- 00:47:45going to go into a a sort of like
- 00:47:47nonprofit trust that um is sort of
- 00:47:49responsible for helping it mature into
- 00:47:52itself um and uh you know gradually get
- 00:47:56smarter and and gain more and more
- 00:47:58Independence um and a few other like
- 00:48:01drops com in just more of a deep dive
- 00:48:03into how it works and um sharing like
- 00:48:06knowledge lots of Open Source stuff
- 00:48:07we're working on um yeah awesome sounds
- 00:48:11like a like a lot to look forward to I'm
- 00:48:12I'm very keen to see what the actual
- 00:48:14very I'm very tired you could probably
- 00:48:16imagine yeah the the legal the legally
- 00:48:20registered religion is going to be great
- 00:48:22it's the the Church of the Flying
- 00:48:23Spaghetti Monster but instead it's a a
- 00:48:25man's gaping an so there we
- 00:48:30go what have you
- 00:48:32done mate anyway thanks so much for
- 00:48:34coming on you're welcome where can
- 00:48:37people find you if they want to keep up
- 00:48:38to date with you and the truth
- 00:48:41terminal uh probably X is the best place
- 00:48:44um is Andy Erie a y re y um and here
- 00:48:48truth terminal um you know if I get
- 00:48:52disappeared from X um uh or know
- 00:48:57otherwise there's access to my account
- 00:48:58um Google will turn up my my personal
- 00:49:02website which has other places to find
- 00:49:04things I'm working on awesome thank you
- 00:49:06very much oh you're very welcome
- Truth Terminal
- Goats Maximus
- Cryptocurrency
- AI development
- Web3
- Cultural Influence
- Autonomous Systems
- Market Dynamics
- AI Ethics
- Meme Coin