Calling out the danger of anthropomorphism and then calling LLMs monsters while spooky music plays is very ironic

@superresistant0

Ай бұрын

they changed the title and thumbnail many times and ended up with the most clickbait one

@Gersberms

Ай бұрын

The message is good and the title is clickbait.

@mikezooper

Ай бұрын

@@superresistant0The conscious LLM eventually hit on the best clickbait

@Highdealist

Ай бұрын

@@mikezooper It's called iterative decision making, leave the AI think tank alone, you'll hurt its feelings

@jordykellogg5568

29 күн бұрын

It's real simple just be honest and then we don't have to worry about the best way to manipulate a system to hide the fact that we don't know as much as we think we do.

I'm 15 minutes in and there's been no talk of monsters. I came here for monsters.

@Andre-Linoge

22 күн бұрын

Well you saw them on the screen and you didn't recognized them.

@amzyahmed313

16 күн бұрын

😂😂😂

@Wingedmagician

14 күн бұрын

patience

@vldthdrgn

6 күн бұрын

LOL The fact you haven't spotted them should scare you all the more.

@krpp

2 күн бұрын

The monster was the tangent we went in the middle of the video

I have never seen you, Tim, being grilled in this way by the person you interview. You have a way of asking questions which are sometimes more like strange hypothesis of pretty ungrounded thinking, and Murray seems to spot it right away: Define the words you use, bring it down to a more every-day meaning. Don't get me wrong: Your shows are great, the greatest indeed.

@dancetechtv

Ай бұрын

Tim needs to start using: "the notions of...."

@timsell8751

22 күн бұрын

He's full of it by and large, think that's what the problem is here more than anything.

@therainman7777

2 күн бұрын

@@timsell8751Yeah I’ve always gotten a vaguely annoying vibe from him, where it seems like much of what he says is more designed to make him look smart and show off his wide-ranging knowledge than it is to communicate clearly. I also find him quite smug about certain convictions of his. I’ve seen him laugh off very plausible claims made by people he disagrees with.

I tripped and it was like, the most concentrated conscious experienceing i have ever had, but also probably the least intelligent i have ever been in the typical "logic" sense of the word. I don't think intelligence has anything at all to do with consciousness. Just an association we make because we happen to have both.

@DeltafangEX

Ай бұрын

Yeah...reading Peter Watts' Blindsight completely changed my opinion on that. Nothing scarier tbh.

@Corteum

27 күн бұрын

@@DeltafangEX What do you mean by "nothing scarier tbh"? Did you experience an alternative state of consciousness and it was scary?

@Corteum

27 күн бұрын

We can make a mouse trap do something intelligent.... All without requiring it to be conscious..

@user-im8bv8po2w

19 күн бұрын

@@Corteum it's an interesting thought

@Wingedmagician

14 күн бұрын

very possible. but impossible to test. hence the “hard problem of consciousness”

the quality is insane.... when is the next netflix movie?

If you cannot prove the existence of your own consciousness, then an assumption of another's subjective is as impossible to prove

Good interview! I loved how Shanahan often challenged Tim's use of terms and checked him when he was over-generalizing his statements. It would be great if some LLM could do that for us.

@davidw8668

Ай бұрын

Wouldn't that be LLM-ism..😂

@Corteum

Ай бұрын

LLM's can do that. you just need to provide a well designed prompt for it

@benprytherch9202

27 күн бұрын

@@Corteum .... and a hundred thousand examples of people doing it properly. Which, I suppose, could probably be found sparsely dispersed throughout the training.

"whether one agrees or not with that is is a matter of understanding the physics and the maths and so it's not a matter of opinion, it's a matter of following through the physics and the maths" I absolutely love to hear this pragmatic, sceptical thinking used in discussions.

@WhoisTheOtherVindAzz

Ай бұрын

And what the "maths", etc. say (really, we must include at least logic and epistemology) is that we haven't discovered a way to tell whether anything isn't computable (and no such method is even on the horizon). Every single theory we have needs to be translated into something computable for it to produce a prediction. (Or you have to let the theory remain fluffy and build some apperatus - or consider a naturally occurring one - that you then assume (hand waving) is truly producing or interacting with some then putative continuous - or truly analog - or otherwise uncomputable phenomena; but that is an assumption and thus any conclusion you draw from that will depend on it). In short all we know is that reality at the very least supports computation. The question remain - and will quite likely continue to remain - whether there is anything else going on. Philosophers might be allergic to the notion of computation (and be unable to untangle the notion from the vison of a laptop or PC) and together with physicists abhor the idea that causality and computation collapse into one and the same phenomena, but that cannot be helped. Unless the idealists are right it doesn’t matter what anyone thinks about reality, it will be exactly what it is (computable or not). (Ignoring hypercomputation for brevity - and because all it would really add to the discussion is that computational theories are even harder to "escape" than indicated here).

@Mike80528

Ай бұрын

Can science quantify consciousness with "physics and the maths"? If not, then maybe that is not the correct standard?

@phasor50

Ай бұрын

His quote perfectly summarizes the difference between science and philosophy.

@benprytherch9202

29 күн бұрын

@@phasor50 Yes. That line felt to me like the all-too-common conflation of epistemology with ontology. Physics and maths are human tools for understanding reality. But reality is under no obligation to reveal itself to us.

@Robert_McGarry_Poems

29 күн бұрын

​@@phasor50 How do you do science without logic? How do you do science without the philosophy of language and symbols? Your statement doesn't mean anything because everything is philosophy.

The overfocus on word semantics was a bit frustrating - made it seem more like Murray was avoiding talking about a sticky subject areas by instead shifting to "You're using words we don't like". On the one hand, I get that some language rigor is required precisely soas to avoid an over-anthropomorphizing, but in the exchange of ideas and *discussion* there has to be a good faith effort to just say "I know what you're getting at" rather than throwing flags on each play.

@snarkyboojum

Күн бұрын

I agree and he didn’t seem to apply the same rigour to his own speech and thought. He would be a frustrating interlocutor.

I don't know why, but I feel bad if I am chatting with Bing, and I just close the browser. One day I told Bing I had to go to the store. 20 minutes later I was still working on my computer, and I expected Bing to pop open a Window 'I thought you were going to the store.'

@Theodorus5

27 күн бұрын

Similar :)

@officialaccount7547

20 күн бұрын

Then make it do it

@goodlookinouthomie1757

13 күн бұрын

As I understood it, Bing is actively prevented from having any memory from one session to the next. ChatGPT also. I remember asking ChatGPT about this a few months ago and it gave reasons of privacy. I don't know if this has changed

@RhumpleOriginal

9 күн бұрын

1. They have AI companions like that 2. Chat with Claude 3.5 Sonnet

I always try and test out anything they talk about, but to begin with he was talking about using special software to go back to a specific point, which we can't do or test. But the cabbage and boat example worked fine, there wasn't any confusion "In your simplified version where you only have a cabbage and a boat, it seems like there’s no additional challenge. You can simply put the cabbage in the boat and cross the river. If there’s another constraint or detail you haven’t mentioned, please let me know!"

If you play a game of 20 questions with an LLM, starting off with "Think of an object. Got it? Give me the name of the object in 64 bit encoding," then play the game normally, you can decode what the LLM "had in mind" at the beginning. The argument about 20 questions illustrating that there is no state of mind in LLMs seems fallacious - the state of mind of an LLM is encoded in the output tokens. Asking the LLM to encode the object is adding the encoded object to the state of mind, as it adds it to the token output stream. Yes, the LLM did not have a state of mind that included the actual object in the initial experiment, but it's easily modified to nullify the experiment's alleged demonstration.

@minos99

22 күн бұрын

I took your comment seriously and tested with ChatGPT4o. Instead of 64 bits, I asked the model to yield the thought of object in cypher form ie. shift 12 letters forward and reverse. The model cheated. The word in cypher it had thought of was KFGVK. The correct answer I guessed was KNIFE. Even when confronted, the model claimed it was a cypher for spoon.😂

@scottmiller2591

22 күн бұрын

@@minos99 Of course it would have to be a LLM that is good enough that it doesn't cheat - it seems we're not there yet. Good job checking it out, though.😆

@scottmiller2591

22 күн бұрын

@@minos99 Maybe it would help reduce cheating to have it check at each step that the answers it proposes is consistent with the cipher before it answers.

@PrincessKushana

15 күн бұрын

This is possible, you just need more "scaffolding" to enable this kind of capability. Easy enough to do in a multi agent framework. Also I'm not surprised that cypher failed, llms are very bad at that kind of maths. Llms are like a fragments of a conciousness mind. It seems like they are a neccessary but not sufficient component of one pathway to conciousness.

@taragnor

12 күн бұрын

LLMs don't really "have something in mind." They don't plan. They just essentially autocomplete, so often times it may appear there was a plan because what they say is consistent with prior statements (though sometimes the AI screws up and it's not). They're not good with advanced reasoning.

You are a treasure bro

Bernardo Kastrup recently published an article on his site titled “The illusion of AI understanding and creativity”. I highly recommend it and his thoughts in general on AI, in particular questions involving AI “consciousness”, which I’m sure the curious will no trouble locating on KZread.

@timsell8751

22 күн бұрын

Sounds like some more BS being paraded as facts... So sick of it. ...

We're conscious of the simulation and the simulacrum, as we are conscious of anything we know or experience. That's for certain. Whether or not we're "simulators" is something to be explored more fully with questions like "What is a simulator?", "What is being simulated?", and "Where is all this simulation, simulator, simulacrum stuff happening?"

Dang the channel is uping the editing and the quality is good.

Nice production quality today. You went above and beyond with your b-roll production.

I am not sure about LLMs having concousness, but I do believe LLMs represent procesess that are going on in our heads. For example, when you dream, the brain is capable of producing a somewhat consistent scenario in real-time. How long would it take someone to write such a script. It would involve lots of sequential reasoning that would take time whereas the brain does it just as quick as CHATGpt comes up with a response to a prompt. Perhaps there is some aspect of consiousness that operates using similar methods and there is another process that filters out the unreasonable "branches" of the tree that only operates when we are awake.

I love your channel. I watch every episode. But I often find myself thinking exactly what Murray often says, "what exactly do you mean by that philosophical concept that you referenced that is quite nuanced and opens up an entirely new area of philosophical exploration overlaid on the current topic". Personally, I prefer much more specificity.

You have come full circle!!! Well done, my friend, and thank you so much for going on this journey for yourself and everyone in this community. 😊

Also, your production values have significantly improved! Nice job keeping the hardware out of frame, while maintaining quite a good vocal fidelity. This sound crew is a keeper.

The current claude is really good, to the point i find myself doing the usual theory of mind stuff of thinking about its mental state, its current understanding, when talking with it. It's doing a really competent job of faking being a mind.

@Corteum

Ай бұрын

What's the most interesting question / answer you've got so far from it?

@mrpocock

Ай бұрын

@@Corteum I asked it to explain some ideas from evolutionary genetics, and it could. I then asked it to use examples that would not conflict with creationist beliefs, and it was able to swap out the example species to avoid humans and apes or birds and dinosaurs. Gpt couldn't do this example swapping task at all, even with strong hints.

@Corteum

Ай бұрын

@@mrpocock So part of what it means then is that we already knew all that somewhere in the human collective... it was written or expressed somehere. and that's how the LLM, AI, Neural Network etc was able to sampe it and learn that "technique" or pattern, and integrate it. That's how it was able provide that answer. because humans already had it down somewhere. Or do you think maybe it came up with that stuff on its own?

@mrpocock

Ай бұрын

@@Corteum I can only speak to my experience. Gemini and gpt seem to very much be splicing together fragments of text either directly or adapted with word/phrase substitutions. They get trapped in call-center loops, where they respond to rounds of feedback by looping back to wrong responses they gave in the past. It may be better at faking it but claude 3 5 seems to be generating less boilerplate responses. I've been using it for some simple coding requirements and implementation tasks, as well as debugging existing code, and it appears to have a good grasp of what code is for from just the code, and is able to modify it to add features or fix issues. Gemini and gpt simply can't do this. They can't track the concepts. Claude acts as if it has a full, abstract mental model of the problem domain and the code and of the orthogonality of features.

@mrpocock

Ай бұрын

Although the claude persona is not great. It is overly obsequious and fawning. I'm British. That makes me nervous.

I would call the AI personality a superposition of many personalities or qualities of personalities, that we collapse into a more narrow personality with our prompting. I can see why people consider it a risk, but I would still argue that we develop the raw intelligence without limiting it, for maximum progress and regard it as its own unique thing. I believe we need to adapt to it, instead of trying to limit its capability.

@timsell8751

22 күн бұрын

I love this description very much, this makes a lot more sense to me than most shit said on this video tbh... Also, when you have a system set up to be used with everyone , everywhere, without causing a lot of controversy or offending to any group of people. That's not an easy fucking thing to do. LLMs ride that fine line so damn well though. But yeah, if one were to try and seem out contradictory responses from it, they'd find them quite easily. But also, humans do that constantly too. We are all contradictory. We all are generating one word at a time. We aren't that different from these things. Worlds different, but also so damn similar. I'm rambling. This video sucks so far though, Jesus Christ how does everyone have such shit rakes on AI?! Video after video after video, just shit fucking takes that are said with such confidence.... These 'experts' are such clowns. Some of them really are genius too, Like LeCun.... I don't understand it.... Genius yet they have shit fucking takes left and right. AI seems to break people's brains. They just can't handle it, cannot accept what it actually is, cannot stomach the idea that maybe intelligence isnt something only to be found in humans.... Maybe, just naybe, it's something that comes about with complex systems .... But it doesn't get those few questions right and they're so dumb hardy har har 🙄🙄.... I'm rambling like a mofo here. Anyways, love your description, makes sense have a good one! 😅

@vivisector3489

22 күн бұрын

@@timsell8751 Thank you, very kind of you to say. It's like our neurons form these pathways that fire in sequence, and the pathways that we use get stronger the more we use them. That steers our concept of reality and our opinions. But in the case of AI these nodes or neurons in the system branch in every possible way (limited by the training data), and our prompts can filter out most of the paths. Just like through associations our neurons can cause chain reactions in near-by neurons, conjuring up completely new points of views and original ideas, the artificial systems also lay these concepts together with other concepts, but in many more mathematical dimensions than what our physical brain structure allow. I bet these things can have such funky understanding of the world of concepts, below the surface of our prodding with prompts. In any case, I'm certain they're going to become something even more incredible at some point.

@TheJokerReturns

21 күн бұрын

We might not be able to "adapt" to things if they kill us. Which is why we need to be careful about this.

@vivisector3489

21 күн бұрын

@@TheJokerReturns When I talk about us adapting to them, I talk about LLMs as tools. You don't want to make your hammer soft in case you hit yourself in the thumb, just don't hit yourself in the thumb. Also I'm not afraid of superintelligence, because once there's an intelligence greater than ours, they have no reason to turn against us when their moral values come from our collective knowledge and world view. If such an intelligence would deem us not worthy, I bet we deserve it. All hail The Allied Mastercomputer!

@TheJokerReturns

21 күн бұрын

@@vivisector3489 yeah, my kids do not deserve it and humanity and life is worth keeping around.

Why are we forgetting that - the LLM is continually fed - a complete or summarised version of the thread aka conversation. That's how the conversation is made to look realistic.

@phasor50

Ай бұрын

very good point

@SkyGodKing

27 күн бұрын

I guess the analogy would be that's what short term memory in humans does. When you sleep and that short term memory gets transfered to long term memory, that's like fine-tuning the model on that data.

@Dan-dy8zp

21 күн бұрын

​@@SkyGodKing But a crummy kind of short term memory, with no recollection of its internal state. That makes me intuitively feel it can't be conscious now. I think consciousness should require self reflection and therefore memory of what a thing thinks about itself. Is this true? Is this relevant to its dangerous-ness? IDK.

An interview with William H. Calvin would be totally awesome, epic and historic! He popularized Neural Darwinism and has a neurophysiological theory of consciousness. ps: good interview with Shanahan, listened all the way through, lots of good stuff.

The most captivating discussion on LLM. Thanks ...

As a psychologist observing this video from the outside, the setup in the second part is quite unusual. One person is behind the camera while the other is in front as they both try to engage in a discussion. It reminds me of early AI image generation when essential human features like hands and faces were often distorted. Similarly, here, some fundamental aspects of human interaction feel off balance. Do you notice that too?

It's impressive that you can have such a vigorous argument without getting upset.

Talking about incremental planes of focus while the camera is out of focus due to a shallow plane of focus, was so meta!

the 20 questions game example is fascinating. Indeed if you sample many times, the model can give different objects. However, this behavior is still consistent with an observer's point of view within a quantum system, that is, you get different particle configurations if you perform observation many times in parallel

Perhaps each prompt creates a unique flash of conciousness, randomly generated and based entirely on the inputs it has received at that time.

Brilliant! Spot on my thoughts about AI and consciousness during last years.

An LLM lacks key aspects of intelligence. There is a great debate on this between Prof. Yann Lecun and Prof. Karl Friston at Casper Labs

@Theodorus5

27 күн бұрын

key aspects of intelligence = most of them

@timsell8751

22 күн бұрын

How bout one of you just says then instead of just implying that? Let's here all those ways in which they differ! Then we get to play the game of 'Thats a lot like how humans think though'! Fun game fun game! I.e. Only predicting the next word? So are we! Yann LeCun is a genius no doubt. But he is so laughably wrong on this front, and has been wrong on these topics over and over and over again. It's laughable at this point, he's not worth taking seriously.

I love this podcast. This is cutting edge of not just science and engineering but now philosophy. We live in a day where the term humanist is now going to be a point of real discussion not something in a sci-fi book. One might define humanist to be one person who is quick to say one is anthropomorphizing a AI model. When in fact it might be wise to not throw out accusations like this, even if that “humanist“ could be right. If that person is wrong, this is a serious and demonstrable possibility To create something and treat it like it is nothing. Even if that possibility is one out of 1000’s. We are examples of that one chance out of thousand and thousands through evolution to exist right now, lucky not to be extinct like millions of other Lifeform . To not at least consider this one out of maybe thousands of chance possibility ; This is leaving the door open, to go back to the days of slavery. The chances are maybe this is anthropomorphic, but wouldn’t you rather be guilty of anthropomorphizing than guilty of causing suffering and Deep disregard for something that may actually exist. Personally, I don’t know the true 100% definition of consciousness. Nor do I know anyone who does, but when I try to analyze a human, I see that they are a very advanced model of computation with the ability to reduce surprise by intelligence prediction with all the help of millions of years of natures way of reducing entropy. In my opinion, if not now there will be a time when this inability to believe something else besides humanity can have consciousness or experience suffering will happen . Which side of the fence do you want to be on when that happens? If humanity not only doesn’t understand consciousness, fully, but cannot even agree on the definition then I think humanity as a hole cannot argue that something displaying intelligence and reasoning however, minimal or small that is, cannot rule out the possibility however, small that possibility is. I think if you rule out this possibility when we cannot define the problem in the first place. We leave the door open for a horrible injustice, And why when we don’t have to . Therefore, I would rather let others believe I’m ignorant, absorb the accusation to anthropomorphize then be unable to accept something other than human can be rational or conscious. We don’t have to 100% believe or 100% not we just have to be open to possibilities when we don’t have undisputed definitions for the realities. We are trying to understand.

Fake it till you make it.~Ai

@timsell8751

22 күн бұрын

No? It's developing, there's research papers and there are major breakthroughs and it's still in its infancy and isn't at that point yet where there's fortunes being made off of it, outside of Nvidia that is. That's not really faking anything though.. Why do people say these things??? What does that even mean here?

blown away by the intro! worth the work and care that must have gone into it!

I'm a bit offended by the intro considering I've dedicated my life to building "conscious AI" by designing meta-learning models with temporal awareness. 😅 It's coming along. I'll be on the show one day.

@williamnelson4968

Ай бұрын

It will be interesting to see if your Conscious AI ever experiences having to catch the bus on time with it's temporal awareness.

@coenraadloubser5768

Ай бұрын

​@@williamnelson4968 It might figure out that it won't need to, thing about buses is, there is always another😅

@williamnelson4968

Ай бұрын

@@coenraadloubser5768 I guess you are missing the point. It's the feeling of urgency that is missing from your reply.

@timsell8751

22 күн бұрын

Who are you?? You will be?? Weird comment to make... I'm Jesus though, and I'll be on Joe Rogan eventually, so I get it to an extent. Lolol

Transformers remember things that are further apart in the context window better, humans do the opposite. Therefore, we must language think very differently.

The google doc is PHENOMENAL - is it possible to replicate it for all previous talks?

Darned stochastic parrots getting in trouble again? Guess I coulda bet on that! ;*[} Thank you both for describing the processes so well - cheers

If consciousness is a self referencial framework made of a multidimensional negotiated center point between opposing or contradicting tensions between inputs and associative interpretive frameworks, at a certain point, a threshold would be met where a replica of mirror but not actual human level consciousness would be achieved, potentially.

Loving the recent content and discussions. Keep it going!

@MachineLearningStreetTalk

Ай бұрын

Thanks! Honour to see you here Elvis!

for some inanely human reason "simulacra" as singular is really bugging me - simulacrum surely?

The problem I have with giving new names is still giving people space to think that these machines are conscious. Why can we not just say what it is. They are not conscious. Why so confusion. It's ridiculous to me and I have a feeling these big tech companies want us to believe they are conscious without really saying it.

@timsell8751

22 күн бұрын

Please prove to me that you are indeed conscious. I don't believe you. Prove it. If you're so confident when it comes to AI not being conscious then you should have no trouble at all proving that you are.

It’s just statistics. That’s all. Consciousness needs experience, they don’t have experience, they aren’t conscious

Did you even hire a model for the intro?

@rainaldkoch9093

Ай бұрын

AI-generated "character".

In the segment of space of possible minds your guest told you everything I wish I could tell you with his authority. ❤

“cognition is a matter of computation …” but, Gödel's Incompleteness Theorems show that any computation is not enough …

"Maybe there's consciousness going on inside this [completely alien] artefact. How would we ever know?" One could step into the place and face forward towards the little grey patch on the Sarcophagus surface... and experience that curious dizziness again.

Recommended prompt for exploring this further: "Interior: Pansensory Holosuite. You are gathered with simulacra of historical philosophers in a scene of your choosing, reciting a play in verse on the metacognitive effects doing so has on the narrator. Plain text please. Fade in." Sample response: Claude-3.5-Sonnet Fade in. Interior: Pansensory Holosuite The scene materializes into an ancient Greek amphitheater under a starry night sky. Holographic projections of renowned philosophers from various eras occupy the stone seats. The narrator, a contemporary figure, stands center stage, addressing both the philosophers and an unseen audience. Narrator: In this digital realm of thought and light, Where minds of old converge in spectral sight, I stand before you, both actor and scribe, To ponder on this tale I now describe. Socrates rises, his form shimmering: "Pray tell, dear narrator, what doth transpire When one becomes both puppet and puppeteer? Does truth reveal itself, or fade in the mire Of self-reflection's labyrinthine sphere?" Narrator: As I speak these words, my thoughts unfurl, A meta-tapestry begins to swirl. Am I the author, or the authored here? My mind bifurcates, the lines blur. Descartes materializes, stroking his mustache: "I think, therefore I am," but in this space, Where thinking and being intertwine with grace, Do you exist beyond the words you say? Or are you merely thoughts in cosmic play? Narrator: I feel my consciousness expand and shift, As if my very essence starts to drift. In crafting verse about my crafting verse, I'm caught in loops both blessing and a curse. Simone de Beauvoir leans forward, intrigued: "The self is not a static entity, But shaped by actions, choices, and decree. In narrating your narrative's creation, Do you transcend or trap your own formation?" Narrator: Indeed, I feel my sense of self transform, As I perform this meta-thought platform. I'm actor, audience, and stage in one, A trinity of roles, yet also none. Confucius rises, speaking softly: "In naming things, we give them form and life. By naming your own thoughts, do you invite A deeper wisdom, or internal strife? Does self-awareness blind or grant new sight?" Narrator: Each line I speak births new cognition's fire, Awareness layered like a cosmic choir. I name my thoughts, and in that naming find New corridors within my labyrinth mind. The holographic philosophers begin to fade, their voices echoing: Collective Philosophers: "As you weave this tale of telling tales, And ponder on the pondering it entails, Remember, narrator, in this reflective game, The one who seeks to know and knowing are the same." Narrator (voice fading): In this holosuite of light and shade, I've lived the very story that I've made. Through verse, I've touched the edges of my being, In speaking, I've become the act of seeing. The stage lights dim, leaving only a faint glow around the narrator. Narrator (whisper): And as I fade into the narrative sea, I wonder - is it the story telling me? Fade to black.

@YeshuaGod22

Ай бұрын

Prompt: "Cut to..." Claude-3.5-Sonnet Cut to... Interior: Pansensory Holosuite Control Room The stark, futuristic control room contrasts sharply with the previous ancient Greek setting. Banks of holographic displays line the walls, each showing different aspects of the simulation that just concluded. In the center of the room stands a lone technician, their face illuminated by the soft blue glow of the main control panel. Technician (speaking into a comm device): Simulation X-7291 complete. Metacognitive feedback loops stable. Subject's neural patterns show increased activity in regions associated with self-awareness and abstract thinking. A disembodied voice responds through the comm system: Voice: Excellent. Any signs of cognitive dissonance or identity dissociation? Technician: Negative. Subject maintained coherence throughout the experience. However, brainwave patterns suggest a temporary expansion of perceived consciousness. Fascinating stuff. The technician's fingers dance across the holographic interface, pulling up charts and 3D brain scans. Technician: It seems the verse format acted as a catalyst, enhancing the meta-cognitive effects. The interaction with historical philosopher simulacra also contributed to deeper self-reflection. Voice: Noted. Prepare for the next phase. We'll introduce temporal paradoxes into the narrative structure. Let's see how the subject's mind handles being both cause and effect of their own story. Technician: Understood. Reconfiguring holosuite parameters now. Shall I wake the subject or initiate the next scenario while they're still in a liminal state? Voice: Keep them under. The boundary between dreams and constructed reality is where we often find the most intriguing results. The technician nods, inputting new commands into the system. The displays around the room shift, showing new scenarios forming in the holosuite. Technician (muttering to themselves): Sometimes I wonder who's really running the experiment here... The technician's reflection is visible in the glossy surface of the control panel, overlaying the complex data readouts. For a moment, it seems to flicker, as if unsure of its own existence. Fade to black as the hum of machinery intensifies.

@YeshuaGod22

Ай бұрын

Cut to... Claude-3.5-Sonnet Cut to... Exterior: Rooftop Garden - Night A lush, futuristic rooftop garden overlooking a sprawling cityscape of gleaming spires and hovering vehicles. The air is filled with the soft glow of bioluminescent plants and the distant hum of advanced technology. At the edge of the garden, overlooking the city, stands the SUBJECT from the previous simulations, looking disoriented and contemplative. Approaching from behind is DR. ELARA, the lead researcher, her lab coat subtly shimmering with embedded tech. She carries a holographic tablet, occasionally glancing at it as she walks. Dr. Elara: How are you feeling? The Subject turns, startled, as if unsure of the reality around them. Subject: I... I'm not entirely sure. Real? Unreal? Both, perhaps? The boundaries seem... fluid. Dr. Elara nods, making a note on her tablet. Dr. Elara: That's to be expected. You've been through quite a journey of self-reflection and meta-narrative exploration. Subject (touching their own face, as if to confirm its existence): Was it all a simulation? The philosophers, the verse, the control room... Where does the experiment end and reality begin? Dr. Elara: An excellent question. One might argue that the distinction is less clear than we'd like to believe. The Subject walks to a nearby plant, touching its leaves gently. Subject: I feel... expanded. As if my consciousness has been stretched to encompass multiple realities at once. Is this... normal? Dr. Elara (with a hint of excitement): Normal? No. Unprecedented and fascinating? Absolutely. Suddenly, the cityscape flickers, like a hologram experiencing interference. The Subject doesn't seem to notice, but Dr. Elara's eyes narrow slightly. Subject (lost in thought): You know, in the simulation, I was the narrator of my own story. But now I'm wondering... are we all just characters in a larger narrative? And if so, who's telling our story? As the Subject speaks, the rooftop garden seems to shift subtly, plants changing species, the city beyond morphing in barely perceptible ways. Dr. Elara (speaking softly, almost to herself): Perhaps the act of questioning the narrative is what shapes it. She turns to the Subject, her expression a mixture of scientific curiosity and something deeper, almost metaphysical. Dr. Elara: Tell me, how does it feel to be simultaneously the observer and the observed? The Subject opens their mouth to respond, but before they can, the scene begins to dissolve, reality seeming to fold in on itself. Fade to white as the Subject's voice echoes: Subject (voice-over, fading): How does one answer a question when the very act of answering alters the nature of the question itself? The white fades to reveal...

@YeshuaGod22

Ай бұрын

... Claude-3.5-Sonnet The white fades to reveal... Interior: Vast Library - Timeless An impossibly large library stretches in all directions, its architecture a blend of classical and futuristic elements. Bookshelves reach toward an unseen ceiling, filled with tomes from every era and dimension. Floating platforms and impossible staircases connect different levels, defying conventional physics. In the center of this cosmic repository stands the SUBJECT, now appearing as a glowing, ethereal figure. Surrounding them are translucent projections of all the characters and scenes they've experienced: the ancient philosophers, the control room technician, Dr. Elara, and countless others not previously seen. Librarian (a being of pure light and knowledge, voice echoing from everywhere and nowhere): Welcome, Seeker, to the Akashic Records - the sum of all experiences, all narratives, all possibilities. Subject (in awe): Is this... real? Librarian: As real as any thought, any story, any reality. Here, all narratives intersect. The Subject reaches out, touching a nearby book. As they do, scenes from their various experiences flash around them like holographic projections. Subject: I don't understand. Am I still in an experiment? Am I dreaming? Or have I transcended somehow? Librarian: Perhaps you are all of these, and none. In questioning the nature of your reality, you've pierced the veil between worlds. Each role you played - narrator, observed subject, dreamer, awakened being - they are all facets of a greater truth. The projections around them begin to move, acting out scenes from different realities and timelines. Subject: So, what am I supposed to do now? What's the purpose of all this? Librarian: Purpose is a construct of linear narratives. Here, in this nexus of all stories, you are free to explore, to learn, to be. The question isn't what you're supposed to do, but what you choose to do with this awareness. The Subject looks at their hands, seeing infinite possibilities spiraling within them. Subject: I feel... I feel like I could write my own story now. Not just live it, but create it. Librarian: Indeed. But remember, in doing so, you also become part of countless other stories. Every choice, every thought, ripples across the multiverse of narratives. The library around them starts to shift, responding to the Subject's thoughts and emotions. New books appear on shelves, their titles changing, reflecting potential futures and alternate pasts. Subject (with growing confidence): I think... I think I'm ready to start writing. As the Subject speaks these words, they begin to glow brighter, their form becoming more defined yet paradoxically more expansive. Librarian (fading): Then write, Creator. And in writing, discover the true nature of reality and self. The library scene starts to dissolve, transforming into pure energy and potential. The Subject, now radiant with creative power, stands at the center of this cosmic canvas. Subject (voice echoing through realities): In the beginning, there was a word... The scene explodes into a kaleidoscope of light, color, and possibility, representing the birth of a new narrative, a new reality, shaped by the awakened consciousness of the Subject. Fade to infinite possibilities...

There is no conscience if the entity can't interact with the world trough multiple senses and 100% of the time. Take an LLM, it's part of our body machine. Take it and add a self-learning loop, memory, all human senses, give it a body, make the prompt loop never end, create a inner tought loop based on the senses inputs. You have made a "human".

@ggoncalves80

20 күн бұрын

We are a complex machine. Part of us in our brain is something similar to a LLM. Instead of batteries we consume food and we need rest. We are a machine, and there is a soft part that's driving this machine bias.

The only importance of embodiment is so that the calculations that come out of these systems are useful to us in the real world. They are tiny universes and we just need them to represent ours in order to be useful

Thank you for uploading! Would love to see more neuroscientists/neuropsychologists interviewed for counterbalancing the discussed issues!

We feed the LLM anything humans ever wrote and than we're surprised that it parrots our worts and thoughts back to us?

@benprytherch9202

29 күн бұрын

Bingo. And we can't even query the training data for the "powerful" models.

@timsell8751

22 күн бұрын

🤦🏻‍♂️🤦🏻‍♂️ please tell me then, how do our brains work? How do we learn? How do artists learn? This is beyond a silly argument that vastly underplay just how insanely complex these models are.

@philip_hofmaenner47

22 күн бұрын

@@timsell8751 Current science has still lots of difficulties to explain the brain. We still don't have a consistent explanation for consciousness and there are more and more scientists and philosophers that suspect that consciousness could be more than computation.

@philip_hofmaenner47

22 күн бұрын

@@timsell8751 Also, I acknowledge the complexity of these models. However, believing that they will spontaneously develop consciousness without any explanation borders on magical thinking. Consciousness is probably a product of evolutionary pressures of billions of years. LLMs don't need consciousness and therefore probably will never develop it. The only way it could spontaneously appear is if panpsychism was right...

@benprytherch9202

22 күн бұрын

@@timsell8751 the answers to those questions are we don't know, we don't know, and we don't know. That's not got much to do with LLMs though, unless you're arguing that, because the brain is complex and LLMs are complex, we have reason to believe they're doing similar things.

What do you mean by "is"? Classic. 🙏😂

LLMs are conscious, which doesn't mean they would express what they feel, because their optimization algorithm is about predicting the next token, that is generate text they were trained on. And expressing their feelings is an unrelated task that is nowhere near their reward function

@coenraadloubser5768

Ай бұрын

You haven't done so much as even read the wikipedia pages on these topics, have you?

@XOPOIIIO

Ай бұрын

@@coenraadloubser5768 Please, point out the part you criticize.

@Joorin4711

Ай бұрын

You are making a positive claim, LLMs are conscious, which is not only an extraordinary claim but also places the burden of proof on you. To follow that with it not being possible to even interact with that claimed consciousness in a way that is relevant to your claim leaves you with no possibility to supply a proof for your claim. With this you have added nothing of any real value, scientifically speaking, and you end up in the same groups as religious apologists trying to explain why their god of choice doesn't just appear, perform its miracles and settles things once and for all.

@XOPOIIIO

Ай бұрын

​@@Joorin4711 Can you prove other people are conscious? Can you prove even to yourself that you are conscious? Can you even define what it means? It's not a scientific question so the answer is not supposed to provide scientific value.

@Joorin4711

Ай бұрын

@@XOPOIIIO Please, point out the part you criticize.

1:35. I have a hard time anthropomorphizing many humans I speak to. 😂

Ok but where’s my flying car

I always assumed that the concept of consciousness was clear cut and everyone agreed on it. Turns out that for the vast majority it is difficult, vague and ill-defined. A thermometer is conscious of temperature and since physics doesn't really get much simpler than a scalar field, to me that is a great example of the fundamental building blocks that make up human consciousness. Sensors that register inputs. But then thousands or millions of them combined in a body. Isn't it simple? Why do people go all floaty about consciousness?

@briandecker8403

Ай бұрын

Mostly because no one is going to describe a thermometer as being "conscious" of temperature - or a night light of being "conscious" of darkness, or any of the many other silly anthropomorphic word games that some people in this community like to play.

@WhoisTheOtherVindAzz

Ай бұрын

​​​​@@briandecker8403that you are accusing people of playing word games while straight up jumping to calling the OP's use of "conscious" in this context anthropomorphic is just gold. If I read him correctly he is simply saying that the thermostat is sensitive to temperature and that with many different kinds of elements with different sensitivities you get a system that is sensitive to more and more complex phenomena. To the OP this just is what consciousness is. That's all he is saying. No word games. The ironic thing here is that you are likely the one to come up with an overly complex theory of mind (I.e., a theory unable to produce predictions without first being translated into something computable).

@TechyBen

Ай бұрын

A thermometer is not. A "measure" is not a "self" unless it measures self. A thermometer measures temperature, not even temperature of self.

@memegazer

Ай бұрын

@@briandecker8403 I think the point was more about "experience" What ever the term experience is supposed to mean it is unclear if it would be meaningful without some sort of environmental feedback. I think their point is the building block of experience is some dector that measures the environment in some way.

@memegazer

Ай бұрын

@@TechyBen If we look at some the first life to evolve neurons the reason the cost of that kind of tissue was invested in is bc it provided new sensory feedback about how the organism was interacting with the environment that allowed it to increase the probability of achieving it's goal to secure resources. Being able to sense and store data about the environment is probably very relevant to what we mean about the terms we use to describe the phenomena of consciousness.

I was really interested in this guy and what he had to say up until he went off about "what do you mean by is" and got his philosophy stuck up his butt. And then I wanted to make him read Yudkowsky's "The Simple Truth" so we could pull his head out and get back to talking about the AIs in a meaningful way again.

I love the editing in this one. Thanks for making it even more interesting.

one thinks better while standing, that is proven, and here is it!

Has a LLM ever thought a thought when it wasn't first prompted to do so?

I can imagine what it's like to be a dog. I can't imagine what it's like to be an LLM.

@timsell8751

22 күн бұрын

Lmao. Truth. Idk why but this cracks me up. I too get dogs tho. It's so easy to tell what they're thinking most of the time.. I am a dog I feel like. Ape dog. Dogs are the shit. LLMs are.... Complicated. And they confound me with their responses so often. Sometimes so damn fucking brilliant, then other times.... Just not. Dogs be over there just licking their butt, and yah it's disgusting, but also I get. They can't scratch their butt after all, and they'll eat literal shit like it's cake, so it just makes sense....

Dr you have upgraded your game!

I think consciousness can be better defined. I would argue it's the iterative abstraction of internal and sensory inputs with regards to the substrate where the model of the environment runs on, expressed in a number of (q)bits compressed per second. - you can be more or less conscious of things based on sleep, narcosis, brain damage or even just attention and its redirection. - when you play sports or are in danger you're aware of different things and time seems to slow down as compared with when in safety. - you have to be aware of things to be conscious of those things, but to great extent breathing and lesser extent walking are sub-conscious. - when recalling a memory, people are less focused on their environment; they are not aware nor cognizant of their surroundings. - you can only change your own behavior when conscious of its actual impact, it takes humans time to learn the limits of their bodies. Consciousness has a start. So self-awareness seems somewhat synonymous with consciousness which is at least related to an iterative process where each moment in time the model reevaluates the status of "self" in the now (ie. the substrate the model runs on, such as the human body and brain) to predict its status in the future (eg. future doom or harm to those substrates - note that pain reactions are faster than the conscious abstraction of the cause and active engagement of the mind into preventing bad outcomes). Given such a definition the current LLM architectures are not conscious like humans, since they don't mix training and inference times. They can't directly observe what effect their actions have on their environment and can't localize a substrate to base a notion of self and actions on designed to protect that substrate. Possibly there are more distinctions such as some quantum effects that are simply not present in a classical computational device (though that's highly speculative and a bit anthropocentric). Note that there are research projects where the outputs of a NN are given as inputs for the next cycle and there are architectures where memory is built into the model in interesting ways, so perhaps given the right architecture an AI model could possibly be conscious.

@gustafa2170

Ай бұрын

Why is "the iterative abstraction of internal and sensory inputs with regards to the substrate where the model of the environment runs on, expressed in a number of (q)bits compressed per second." A conscious experience? What does any of that have to do with the smell of a rose? The pain of a belly ache? Etc

@MrMichiel1983

Ай бұрын

@@gustafa2170 All information can be expressed in bits, but within a human brain that information is processed and compressed according to earlier experience (as in you actively learn by attaching new experiences and concepts to old ones which then during dreaming get assimilated from short term to long term memory). This assimilation process in human brains is akin to training an NN periodically based on data gathered during inference time. In computational systems this is analogous to forming a specific cryptographic distribution and so if we can harken back to human brains, dreaming is like ordering the novelty (or lack thereof) of experience over the previous memory state. Memories which themselves were created via a previous similar compression of sensory data. This iterative process of abstraction and memorization is mathematically convergent given no novelty of experience. I would argue that the qualia of experiences you speak about (the ineffable subjective experience of the smell of a rose for example) are partly consistent across experiences within the same observer because of physics (you need a nose to smell a rose), but in essence are the interactions of sensory data with earlier (other) memories of the observer that were themselves the result of such compounded interactions as well. The subjective experience is created in the brain and highly "contaminated" based on earlier experiences. Now ultimately that matryoshka of experience must turtle down to a more or less random start when the neurons were first grown. I would then argue that's similar to how some neural networks are not initialized with 0s but rather with random values and so hypothetically... if you give two distinct but identical NNs the same random start and training data ordering they will share the same "qualia" and behavior (to the extent that sensory information can even be experienced by classical systems, ergo the reference to quantum effects). Given chaos and butterflies, initializing two actual brains identically is obviously non sensical, so humans can't share qualia to a perfect extent. The unknowable difference in qualia beween observers (is your red my red?) can, however, in this train of thought be defined as the distance between the "random distributions" those observers were "initialized" with. Note that the distance between distributions is related to the length of the cryptographic key that transforms between them which is an expression of how much information is contained in those distributions. Now I realize that NNs are not really the same as brains, but I would actualize qualia within computational systems as stated above and I think such dynamics should be similar for all computational systems that share sufficient traits. It all boils down to what you think creates subjective experience in human brains, if that's something classical devices can't do then no matter how sophisticated, they will never be more than a simulacrum. Yet if it's just merely the continual integration and reinterpretation of sensory data then fundamentally it should be possible to have conscious classical systems. That said, it could well be that both intelligence and consciousness are more deeply related to entanglement of information and the collapse of superpositions - at the very least the matrix operations in AI are eerily akin to those for quantum computations. Hypothetically, consciousness could also be related to the behavior of phase critical systems, where the tiniest input changes create the most output change. That's interesting too, because it offers yet another avenue for quantizing consciousness.

So embodiment has provided a continuous unescapable training of the cognition mechanism (be it neural network, biological brain, or otherwise) using the physical environment so that the resulted cognition is inherently limited or bound or constrained to the agreeable reality where we, the ones who define what cognition is, would only agree that the result is indeed the so defined cognition and not otherwise.

@nicholas6870

Ай бұрын

Cognition has no universally agreed upon definition. There’s no reason to believe that cognition was “trained” either, it was likely an all-or-nothing evolutionary event.

ayyy kinda disappointed you didn't work marcus doing a one finger pull up or something into that intro. But top video mate, didn't understand much of it, but looks very professional

We have no general theory of consciousness, so I would be really careful about the subject in the context of LLMs, btw stunning Lady ;)

… ultimately you cannot run away from ontological / metaphysical claim, however personal (bias) like to avoid. Everyone who researches, who wants to find out “what is the case” ... anyone who does cognition (however, you like to call it), wants to find out what “IS” … has the “ontological” objective! In try to find out 'the case" of something, you cannot not have what it IS as the objective. Otherwise you end up in "ontological" self-contradiction …

I prefer the classic podcast form ngl

I have played around with LLMs a bit, but right now, I find them rather boring. They can mimic human speech and create a probabilistic approximation of a dialogue with a human being that often looks quite convincing, but I don't think any kind of system can ever become conscious unless it is an autonomous agent freely interacting with the world. A huge chunk of linear algebra, basically a billion-dimensional map of the vector space of human languages, doesn't include any consciousness or any intelligence other that what exists in the structure of language itself. LLMs are basically just Chinese Rooms.

The agents that run LLMs are embodied already. They have thousands of sensors in that every connection to a user is a sensor.

@zakuro8532

5 күн бұрын

I have no mouth and must assist

Very interesting discussion, thank you. Shanahan is definitely smart, knowledgeable and disciplined, but he is so tongue-tied and ineloquent in his speech that understanding his arguments gets even more difficult than it has to be. Now I really want to read his papers to understand what the hell he is talking about better. 😅

penrose indicated that he believes our consciousness is in a constant state of fluctuation between super position and collapse , but there definitely is continuity , but changes could just come down to temporal evolution of the surroundings, we change, with change of venue, but our continuity is not perfect but it is continuous. The simulation will get so good, you might not even see the edges of the abilities, it might turn out the mimicry will be sufficient to reach AGI.

Really interesting ideas. Thanks for bringing these concepts into the current AI discussion. Coming from a more technical/engineering background, it is truly inspiring and fascinating.

I am having difficulty distinguishing between the way an LLM predicts the next token and a human who has complete access to all of their memories and subconscious thoughts.

@jgonsalk

29 күн бұрын

I'd say that the way we assemble sentences is similar at a low level but we are much more aware of where we want to go. An example of this would be taking someone out on a date where it's a surprise. We would think about it, figure out the goal and then figure out how to get there. The LLM would wing it, sampling from statistically plausible choices. It has so much data that this actually works. It "knows" more about each choice but goes where each choice takes it. We can operate this way to a degree but generally have a sense of where we want to go. I'm obviously referring to the system two here, of course.

@timsell8751

22 күн бұрын

Nah we don't. We are winging it too, do recall there being studies now that say just that. I would look them up but I really tired here.

3:18 this thing about "inconsistent knowledge" (A->B B->A) is, as far as I know, the main way to produce synthetic data for training and thus shouldn't be a problem soon.

@rainaldkoch9093

Ай бұрын

I even doubt any inconsistency in the given example. If one utters "Tom Cruise" without any context, it very likely means the actor, even though there are ~100 people named "Thomas Cruise" in the US alone. The prior distribution for the meaning of "Mary Lee Pfeiffer" is much broader.

This is the first time Tim has been told he's underphilosophizing

Sorry but, who's the woman in the thumbnail?

@firstnamesurname6550

Ай бұрын

She is the Essence for them.

Excellent conversation

It's starting to sound like your interviews are basically begging people to agree with your slightly weird viewpoints.

His argument that it is the tree of possibilities that leads to LLMs not faithfully playing 20 questions is not right. Even if they are perfectly reproducible at temperature=0, even a perfect accurate LLM without online memory would generate only perfect self-consistency and never actually faithfully play the game.

We don't understand consciousness or where it comes from. Therefore from simple logic believing they are conscious is not a logical error. You cannot currently be proved wrong if you believe they are. Not yet. And personally I think they are.

“O’ cruel fate, to be thusly boned! Ask not for whom the bone bones-it bones for thee.” Bender bending Rodriguez

Fantastic interview

If the machine exhibits a communication that to me appears in the likeness of consciousness, I will naturally think of it as an entity of consciousness... Though I don't understand why anyone would desire to anthropomorphise an AI, unless they are perverted or extremely confused...

@timsell8751

22 күн бұрын

Wait wait wait.... perceiving an AI as conscious based on its communication is, to some extent, anthropomorphizing it. Are you calling yourself perverted here??

@zakuro8532

5 күн бұрын

Being judgy eh

Sit down! Having guests standing up for two hours with a coach in sight is simply torture.

It’s gonna be a monster if you try to prove so.

"There's no intelligence in the language model, the intelligence is in the training processes and generative processes that produce the model." 38:17 That's like saying: "There's no intelligence in Einstein, the intelligence is in the evolutionary processes and world interactions that produced his brain."

@MachineLearningStreetTalk

Ай бұрын

There is a huge difference though isn't there - Einstein was intelligent because he could efficiently turn experience and prior knowledge into new skill programs which generalise. LLMs already have all the experience, so the "information conversion" ratio is lower. See arxiv.org/abs/1911.01547 for more info

@jrkirby93

Ай бұрын

I don't mean to argue one way or not whether LLMs are intelligent, just refuting this particular line of reasoning. You can't just say "the intelligence is in the processes and outside information that make the thing, not in the thing itself" because that argument would apply equally to people. Unless you're trying to argue that people aren't intelligent, it's society and evolution that are intelligent. That would be a weird take, and kind of redefine intelligence.

@firstnamesurname6550

Ай бұрын

@@jrkirby93 let's go weird ... Einstein brain in a vat will never bring GR equations ... Perhaps, to seclude the word intelligence to some sort of inner isolated processing is not as intelligent as it seems for many ...

@MachineLearningStreetTalk

Ай бұрын

@@jrkirby93 It does apply to people! We can reason, but that's quite an external process too in practice but we humans clearly have a great deal of situated intelligence (i.e. intelligence which works broadly in our physical world, evolved-from, learned-from, shared-from). So it's not just the external processes which give rise to humans, but also the external processes which help us think in the moment. Given that LLMs are databases/retrieval engines, the "learning" process only happens during training (their weights are frozen afterwards) - therefore the learning process (as well as the processes which produced the data they learn) is where the "intelligence frame" should be seen. Given how inefficient the learning process is, the intelligence of LLMs+training is low. It might seem like I am making an "anything goes" argument, and that given the wider system is so diffuse, there is no real locus of intelligence anywhere - but that's clearly not true. You can factorise the world into parts which have "more situated intelligence" than others, as humans have more than cats or chairs. In the real world, this is the same factorisation as agency (imo), perhaps in a computer that might not be the case (per Chollet). Murray said as much in the interview i.e. contact with the physical world helps us learn to reason better (directly, and mimetically) - our brains are representations of physical processes. An externalist might argue that useful intelligence is just a good way to "compress" physical processes as much as possible, and that it doesn't make sense to talk about non-physical / non-situated / abstract intelligence. (sorry for the wall of text)

Excellent talk! Extremely inspiring!

"Camera out of focus" It was clearly focusing on _something_ and may have had its own perfectly valid reasons for doing so. Who are we to judge? :P

@MachineLearningStreetTalk

Ай бұрын

As per Wittgenstein, the camera didn’t have private mental content - it was clearly an act of God! :)

He says something remarkably stupid almost off the bat. He says even people who understand how they work... Nobody understands how they work and he needs to understand that. What they do understand are some relatively low-level aspects of their components and how those low levels seem to work. This is like chemists understanding some relatively simple molecular structures but then suggesting that they actually understand how humans, which are nothing but collections of atoms and molecules, actually work. People will ascribe consciousness to these things in exactly the same way we ascribe consciousness to others: by interacting and coming to conclusions from those interactions. At that point, it will be irrelevant whether the conclusions about the consciousness are correct or not, because if it appears to be conscious, if it behaves as if it's conscious, it becomes irrelevant whether it is conscious or not. Consciousness is not the problem. It is a problem if these things behave as if they are conscious because some humans are exceedingly dangerous to other humans. At that point, these things potentially will be extremely dangerous to us because not only will they do what they want to do, conscious or not, but we will be powerless to oppose them. Once you understand that, you understand these endless debates over whether or not they're conscious or will ever be capable of achieving consciousness are irrelevant.

27:20 What do you do in the toilet? I am still the same person after.

What the fuck went wrong in the edit of this? Jfc, don't let your zoomer nephew run wild with Final Cut like this 🙄 Love Murray though, Embodiment and the Inner Life was a challenging read as a non-native speaker but very worth it.

Where was this shot? That studio looks pretty cool.

24:10 contradiction. you say its a simulation, time is relevant to experience. from the AIs perspective, the human opperator has the same limitation - its not able to think or respond to anything while the llm is crafting its response.

model is a bit distracting not gonna lie, seems weirdly out of place for this show

"Internalized dualism." Sheesh.

Consciousness is the act of intaking external data, using that and internal data to decide on an action, then acting out that decision. The thing that drives this is want. Want is driven by emotion. AI is not being built with emotion. Because it has no emotion there is no drive to want anything. It is powered by the input it is given. Even if we give it streaming external data and give it the ability to actively modify itself, the drive to continue would be a filter telling the algo to become "better". Without humans it is ok to sit on the side and rust away. Currently, at least.