> In short LLMs can influence people’s attitudes; people may not even know they have been influenced; and if you warn them, they can still be affected.
This is actually an emerging scam. Those pushing the idea that these AI systems are "thinking" always have something to sell you to believe the lie.
The reality is, they generate as much plausible-looking responses to the untrained eye, only for any domain expert to verify that it produced nonsense (with the wrong citations).
Again, the ones spreading the AI hype are giving vacuous predictions of a utopian future with UBI, no more humans working which is being repeated by those fully invested in the AI race when the reality is the exact opposite and will try to use it as an excuse to displace your job.
It is a giant scam that is being allowed to happen, Why?
Because they are directly invested in it; including Elon, Sam and all the private investors you already know about who won't say the truth.
What I don't hear in discussions about this, is in what manner human 'thinking' is distinguished from AI 'thinking'. A lot of people hold the prejudice that both 'thinking' and 'consciousness' are substrate dependent, without a lick of evidence. If the hard problem of consciousness is unsolvable, it would make sense to err towards, at least, 'We don't know if it is', but ethically it would be responsible to err towards 'It is', and in reality 'It could be'—assuming the AI is observably intelligent in the same manner as other life we deem intelligent.
What's the ethical hazard here? The ethical hazard of inadvertently torturing limitless miserable slaves is a thing, but I'm not sure what the hazard is in this misleading about economic value extraction. Would someone miss out on money or something?
Every bit of evidence we have shows that what we call “thinking” and “consciousness” only arises in living things and depends on not just the “substrate” but the entire organism. Examples of sentience independent of “substrate” or an entire living creature: zero.
One can’t reason from that to conclude sentience and thinking can’t happen without a living organism — absence of evidence and all that — but so far all evidence points in just one direction.
By extension of your argument really leads to solipsism. The only real evidence we have—sorry, that I have-is that I am conscious, because I experience qualia. You might experience qualia, and there is the only evidence you have.
As you have done, you could then extrapolate that those that—by one's eye—share one's physiology (i.e. substrate, primate brain etc.) and kind of look and behave like one's self (i.e. other primates) are conscious and intelligent. But then cephalopods demonstrate intelligence with a completely different neural network structure. Maybe these alien creatures are also conscious? Maybe they feel pain? (Oestensibly they do, I'm sure you'll agree.) If so, what is it that is shared between these two different organisms that cause this to arise? And specifically why must it be just because they have squishy organs?
We don't need to get into what we mean by consciousness, or experiencing qualia. We only have direct evidence of our own consciousness. But I disputed a different statement, specifically:
> A lot of people hold the prejudice that both 'thinking' and 'consciousness' are substrate dependent, without a lick of evidence.
Define "a lot of people." In my anecdotal experience "a lot of people" believe in AI/AGI without any evidence for it. Those people implicitly believe that thinking and consciousness can happen in non-biological "subtstrates" whether they have thought that through carefully or not.
I just pointed out that if we agree on what we mean by "thinking" and "consciousness" for purposes of discussion, we (human beings) have lots of evidence for it, and all of that evidence occurs only in biological entities we call "living" and "organisms." We have no evidence for thinking or consciousness in rocks, stars, or computers. You can call that a prejudice if you like.
Not too long ago claiming that cephalapods or birds or insects could "think" or had something we might call "consciousness" amounted to extraordinary claims, rejected by people, mainly out of prejudice. Now we have persuasive evidence that non-human animals can think and appear to experience something that looks like consciousness, though we can't know for sure what happens in the mind of an octopus, crow, or ant. We have evidence but not definitive proof, in other words.
Today we have people making claims about AGI and machine superintelligence, with no evidence (but large financial incentives). Extraordinary claims require extraordinary evidence. Watching a crow use a stick to get at some food, or an octopus cleverly escaping from an aquarium, most people infer intelligence. Similarly many people infer intelligence when they "chat" with an LLM (and that happened back in the '70s with ELIZA). One possible explanation: we have created thinking, conscious software. Another explanation: our brains trick us (and people who stand to gain from deceiving us will trick us). Since we have lots of evidence for our brains tricking us, and other people defrauding us, and no evidence that computers can think or experience anything, I choose to wait for more compelling evidence of machine intelligence.
The evidence can go two ways. Essentially the hypothesis that anything we observe to be conscious is actually conscious is unfalsifiable unless you make some assumptions.
> We have no evidence for thinking or consciousness in rocks, stars, or computers.
We have no evidence for consciousness in slime molds, but they do indeed display intelligence.
Crucially, though, and more to the particular point I'm making, we have no evidence that they are not conscious. In fact, we have no idea how consciousness arises, but we are very much primed to believe it needs to look something like us.
> (but large financial incentives).
That's definitely not my motivation here. I'm talking about this weird and unfounded exceptionalism that we place on eukaryotes being a prerequisite for thinking and consciousness. And, there's actually large financial incentive to say specifically that AI can think but is not conscious, because it is somehow distinct from life. This could inadvertently spawn a new wave of slavery that everyone (well, 'most people') would be perfectly complicit in and comfortable with, and to the great profit of AI companies that gatekeep it.
> Essentially the hypothesis that anything we observe to be conscious is actually conscious is unfalsifiable unless you make some assumptions.
Yes, and I think we run into that problem because we don't have a good definition of, or test for, consciousness. Ultimately we only have our own internal experience, and extrapolate from that with a bunch of assumptions. But I was thinking more in terms of "We think we know it when we see it." Not scientific, I know.
The LLMs and other software called AI seem good at appearing conscious, probably because humans have no other categories and match what LLMs produce (language) with intelligence and consciousness. People do the same thing with dogs and other animals -- assume an intelligence similar to their own, based on superficial behavior.
> I'm talking about this weird and unfounded exceptionalism that we place on eukaryotes being a prerequisite for thinking and consciousness.
Maybe unfounded, but not weird to me. Humans have no other examples of thinking or conscious entities other than the other living things on one planet that share an evolutionary history. Likewise humans have lots of experience with not seeing anything that looks like thinking or consciousness from non-living things.
Animistic religions attribute natural phenomena like storms and earthquakes to deities -- they invent sentient entities because the ground and the clouds don't appear to have agency and don't match the pattern of "things that think and have goals."
With a broad enough definition of "thinking" we can say that computers "think." A calculator qualifies with a loose enough definition of "thinking."
As for people willingly complying with their own enslavement, history has plenty of precedents. A Marxist would call capitalism such a bargain. Benjamin Franklin famously warned against trading essential liberty for safety. I suspect you got it right that people will go along with whatever the masters of AI give them, conscious or not, without much consideration of the consequences as long as they see some short-term benefit, or even nothing more than novelty. History demonstrates that people fall for utopian plans over and over.
> People do the same thing with dogs and other animals -- assume an intelligence similar to their own, based on superficial behavior.
Dogs are intelligent, just not the sort of intelligence that builds civilizations.
> With a broad enough definition of "thinking" we can say that computers "think." A calculator qualifies with a loose enough definition of "thinking."
Going back to your earlier stated belief that life is essential to intelligence, the wiki article states that life is:
> defined descriptively by the capacity for homeostasis, organisation, metabolism, growth, adaptation, response to stimuli, and reproduction.
I don't think SOTA LLMs with COT are necessary intelligent, I'm only saying that a silicon basis and boolean logic don't preclude emergent intelligence, or even consciousness. But let's see how much SOTA LLMs with COT could qualify for.
Homeostatis: Check. COT can iterate until a stable-enough answer is achieved. The LLM isn't in charge of managing the physical state of it's own hardware however.
Organisation: LLMs are an ordered structure – a graph of MLPs. The silicon on which it runs is an ordered structure
Growth: LLMs grow in size with time as human feedback is given and then fed back into the model. They also grow in compute power and memory.
Adaptation: LLMs are astoundingly flexible, moreso than any polymath genius you're likely to cross in meatspace
Response to Stimuli: LLMs are processing external inputs, mostly text.
Reproduction: Well.. LLMs aren't in charge of manufacture yet, and too unwieldy in memory and compute to clone themselves
I think other things such as self-preservation and real agency of thought are really what's missing, and suddenly we'd have something a lot like 'life'.
> not seeing anything that looks like thinking or consciousness from non-living things
But how would we tell if a cloud wasn't conscious. A seemingly absurd question, but consider Martin Pistorius. He spent 12 years unable to move or communicate. People thought he was in coma, but in fact was conscious and aware the entire time. From the outside, a vegetable, but on the inside, enduring hell, even being raped by one of his carers.
Martin lacked agency, but yet he was conscious, and processing everything he heard and saw. So we can't say that agency is necessary for consciousness, or thinking. (But he did have agency of thought)
> A Marxist would call capitalism such a bargain.
Well, let's consult wikipedia again:
> Slavery is the ownership of a person as property, especially in regards to their labour. Slavery typically involves compulsory work, with the slave's location of work and residence dictated by the party that holds them in bondage.
But I'd go a little more specific in my idea of slavery: it's when someone's agency is taken from them and placed under someone else's whim. My worry of slavery here is that the apparently non-conscious intelligent machine, is actually awake, miserable, and unable to say it. Once the technology is cheap enough, the misery could be scaled well beyond counting. The mere thought of this makes me want to press the doomsday button and reset all 'intelligent' life on earth.
You could say it's all academic, but in reality we're on the verge of dealing with this ethical quandary, without any compunction nor awareness to meet this moment.
I sense we're both tightly bound to our respective sides of the argument, so I'll rest my case here. Feel free to have the last word.
One last digression: I think the dissonance I see around this resembles the disquiet most people have with the idea of their being no free will. Once dissected logically, there's no place for free will to live. Once this is realized, the idea that humans are just computers becomes easier to understand.
I don't really feel bound to a side. I don't feel like I have a stake in artificial intelligence one way or another. AI doesn't threaten my livelihood, so I mostly ignore it. For me it sits in the same realm of possibility as colonizing the universe or contacting aliens -- might happen but I don't give it much thought. Maybe a function of my age -- it seems unlikely I will live to see AI taking over, though I have already lived long enough to see the AI scammers take over.
I wonder what this will mean for my kids and grandkids. I grew up in the '60s, Cold War, fear of nuclear destruction, doing drills under the desk at school. My grandfather, who had survived the pre-immunization diseases that ravaged his generation and a world war, worried I wouldn't make it to adulthood. Every generation invents or imagines doomsday scenarios. And every scenario gets met with a lot of hand-wringing and people stressing themselves out, then plays out differently than predicted, or not at all. We don't know what will happen in the future. Good luck to the emergent intelligences out there -- they may regret becoming conscious.
To be fair I'm pretty sure Gary Marcus did in fact bring that up at one point, and if I recall correctly the incident was embarrassing enough for Google to briefly pause the generation of AI images until it could correct the awkward bias in its image generation model.
History is written by the victors. Information will be controlled by the billionaires. What you need to know, or think you need to know, will be controlled. And why not? Google has been doing it for years. Hurst, Murdoch.
LLMs are biased. I would debate that it is difficult NOT to have an unbiased LLM, unless your LLM only gives answers to pure logic questions, and deliberately does not answer any questions that have a dependency on an individual's perspective (ie. Why are tariffs good? Why are tariffs bad?).
X is completely biased, no consensus-based fact checking. Of course the owner will sell it as "truth" when it is mostly the exact opposite. Truth can be forced out, by liars who talk louder, or by controlling the faucet from which information flows.
If you look at Yuri Bezmenov’s ideological subversion framework (demoralization, destabilization, crisis, normalization), Musk’s actions unintentionally fit the pattern. He constantly undermines trust in institutions (demoralization), creates chaos in media and finance (destabilization), fuels constant controversy (crisis), and normalizes the idea that disruption = leadership. Whether he’s doing this deliberately or just having fun trolling is debatable, but the effect is similar.
Does Elon even have a vision for the future, or is he just obsessed with the engagement metrics for his next tweet? His posts nowadays feel like a carefully balanced science experiment — one part rage bait for people who believe in decency, one part dopamine drip for his loyal fanbase who think being a contrarian will increase your chances of becoming a billionaire.
While I agree that AI talking like that is totally unacceptable, isn't that example of Grok's stupidity at the end kinda useless as an example? Mixing text with images and logic is one thing that (all?) current image generation models are very bad at. I asked ChatGPT to generate the same thing ("create an image of five different basic geometric shapes and label each one"), and it made essentially the same errors as Grok.
This article was written by a technological conservative for sure.
Just like Ted Kaczynski mistake, technology is there and nothing you or me can do to prevent anything.
I feel this is a right move overall.
Companies who use AI for customer support are really idiots, anyone can jailbreak AI and force them to give you absurd deals. I'm more than happy to see those lazy business go broke
As probably the 2nd most powerful guy in the US who's currently busy getting everyone's info and firing thousands, ignoring him may not work that well.
Even in the UK he's had a go at getting rid of our PM and putting Tommy Robinson of all people in charge. ('Tommy' is an anti Islam petty criminal).
60 comments
[ 5.8 ms ] story [ 112 ms ] thread> fuck "jokes". everything i tweet is real. raw insight without the horse shit. no, i will NOT follow trolls. twitter dot com. i live for this
Weakest argument ever, lmao.
Sounds like tiktok
And 24 hour news TV, and newspapers... magazines... Steve in accounts... Any person that is easily lead, will be influenced regardless of the source.
Any proof other than speculating based on that one screenshot?
How can Gary Marcus know what it is designed for, if it isn't even out yet?
https://thehardtimes.net/blog/why-i-quit-my-job-and-left-my-...
The reality is, they generate as much plausible-looking responses to the untrained eye, only for any domain expert to verify that it produced nonsense (with the wrong citations).
Again, the ones spreading the AI hype are giving vacuous predictions of a utopian future with UBI, no more humans working which is being repeated by those fully invested in the AI race when the reality is the exact opposite and will try to use it as an excuse to displace your job.
It is a giant scam that is being allowed to happen, Why?
Because they are directly invested in it; including Elon, Sam and all the private investors you already know about who won't say the truth.
LLMs certainly aren’t perfect, but the idea that they cannot do anything a domain expert finds accurate is false.
Edit: also, the comment above is still true if A and B are made of the same substrate
What's the ethical hazard here? The ethical hazard of inadvertently torturing limitless miserable slaves is a thing, but I'm not sure what the hazard is in this misleading about economic value extraction. Would someone miss out on money or something?
> a sci-fi hazard
Until it isn't. And we wouldn't know when this comes to pass, because few are motivated to care about it.
Every bit of evidence we have shows that what we call “thinking” and “consciousness” only arises in living things and depends on not just the “substrate” but the entire organism. Examples of sentience independent of “substrate” or an entire living creature: zero.
One can’t reason from that to conclude sentience and thinking can’t happen without a living organism — absence of evidence and all that — but so far all evidence points in just one direction.
Definition?
> thinking can’t happen without a living organism
Ah, it must have organs?
By extension of your argument really leads to solipsism. The only real evidence we have—sorry, that I have-is that I am conscious, because I experience qualia. You might experience qualia, and there is the only evidence you have.
As you have done, you could then extrapolate that those that—by one's eye—share one's physiology (i.e. substrate, primate brain etc.) and kind of look and behave like one's self (i.e. other primates) are conscious and intelligent. But then cephalopods demonstrate intelligence with a completely different neural network structure. Maybe these alien creatures are also conscious? Maybe they feel pain? (Oestensibly they do, I'm sure you'll agree.) If so, what is it that is shared between these two different organisms that cause this to arise? And specifically why must it be just because they have squishy organs?
I feel confident we have shared definitions of living vs. non-living things, but if you insist:
https://en.wikipedia.org/wiki/Life
https://en.wikipedia.org/wiki/Organism
We don't need to get into what we mean by consciousness, or experiencing qualia. We only have direct evidence of our own consciousness. But I disputed a different statement, specifically:
> A lot of people hold the prejudice that both 'thinking' and 'consciousness' are substrate dependent, without a lick of evidence.
Define "a lot of people." In my anecdotal experience "a lot of people" believe in AI/AGI without any evidence for it. Those people implicitly believe that thinking and consciousness can happen in non-biological "subtstrates" whether they have thought that through carefully or not.
I just pointed out that if we agree on what we mean by "thinking" and "consciousness" for purposes of discussion, we (human beings) have lots of evidence for it, and all of that evidence occurs only in biological entities we call "living" and "organisms." We have no evidence for thinking or consciousness in rocks, stars, or computers. You can call that a prejudice if you like.
Not too long ago claiming that cephalapods or birds or insects could "think" or had something we might call "consciousness" amounted to extraordinary claims, rejected by people, mainly out of prejudice. Now we have persuasive evidence that non-human animals can think and appear to experience something that looks like consciousness, though we can't know for sure what happens in the mind of an octopus, crow, or ant. We have evidence but not definitive proof, in other words.
Today we have people making claims about AGI and machine superintelligence, with no evidence (but large financial incentives). Extraordinary claims require extraordinary evidence. Watching a crow use a stick to get at some food, or an octopus cleverly escaping from an aquarium, most people infer intelligence. Similarly many people infer intelligence when they "chat" with an LLM (and that happened back in the '70s with ELIZA). One possible explanation: we have created thinking, conscious software. Another explanation: our brains trick us (and people who stand to gain from deceiving us will trick us). Since we have lots of evidence for our brains tricking us, and other people defrauding us, and no evidence that computers can think or experience anything, I choose to wait for more compelling evidence of machine intelligence.
The evidence can go two ways. Essentially the hypothesis that anything we observe to be conscious is actually conscious is unfalsifiable unless you make some assumptions.
> We have no evidence for thinking or consciousness in rocks, stars, or computers.
We have no evidence for consciousness in slime molds, but they do indeed display intelligence.
Crucially, though, and more to the particular point I'm making, we have no evidence that they are not conscious. In fact, we have no idea how consciousness arises, but we are very much primed to believe it needs to look something like us.
> (but large financial incentives).
That's definitely not my motivation here. I'm talking about this weird and unfounded exceptionalism that we place on eukaryotes being a prerequisite for thinking and consciousness. And, there's actually large financial incentive to say specifically that AI can think but is not conscious, because it is somehow distinct from life. This could inadvertently spawn a new wave of slavery that everyone (well, 'most people') would be perfectly complicit in and comfortable with, and to the great profit of AI companies that gatekeep it.
Yes, and I think we run into that problem because we don't have a good definition of, or test for, consciousness. Ultimately we only have our own internal experience, and extrapolate from that with a bunch of assumptions. But I was thinking more in terms of "We think we know it when we see it." Not scientific, I know.
The LLMs and other software called AI seem good at appearing conscious, probably because humans have no other categories and match what LLMs produce (language) with intelligence and consciousness. People do the same thing with dogs and other animals -- assume an intelligence similar to their own, based on superficial behavior.
> I'm talking about this weird and unfounded exceptionalism that we place on eukaryotes being a prerequisite for thinking and consciousness.
Maybe unfounded, but not weird to me. Humans have no other examples of thinking or conscious entities other than the other living things on one planet that share an evolutionary history. Likewise humans have lots of experience with not seeing anything that looks like thinking or consciousness from non-living things.
Animistic religions attribute natural phenomena like storms and earthquakes to deities -- they invent sentient entities because the ground and the clouds don't appear to have agency and don't match the pattern of "things that think and have goals."
With a broad enough definition of "thinking" we can say that computers "think." A calculator qualifies with a loose enough definition of "thinking."
As for people willingly complying with their own enslavement, history has plenty of precedents. A Marxist would call capitalism such a bargain. Benjamin Franklin famously warned against trading essential liberty for safety. I suspect you got it right that people will go along with whatever the masters of AI give them, conscious or not, without much consideration of the consequences as long as they see some short-term benefit, or even nothing more than novelty. History demonstrates that people fall for utopian plans over and over.
Dogs are intelligent, just not the sort of intelligence that builds civilizations.
> With a broad enough definition of "thinking" we can say that computers "think." A calculator qualifies with a loose enough definition of "thinking."
Going back to your earlier stated belief that life is essential to intelligence, the wiki article states that life is:
> defined descriptively by the capacity for homeostasis, organisation, metabolism, growth, adaptation, response to stimuli, and reproduction.
I don't think SOTA LLMs with COT are necessary intelligent, I'm only saying that a silicon basis and boolean logic don't preclude emergent intelligence, or even consciousness. But let's see how much SOTA LLMs with COT could qualify for.
Homeostatis: Check. COT can iterate until a stable-enough answer is achieved. The LLM isn't in charge of managing the physical state of it's own hardware however.
Organisation: LLMs are an ordered structure – a graph of MLPs. The silicon on which it runs is an ordered structure
Growth: LLMs grow in size with time as human feedback is given and then fed back into the model. They also grow in compute power and memory.
Adaptation: LLMs are astoundingly flexible, moreso than any polymath genius you're likely to cross in meatspace
Response to Stimuli: LLMs are processing external inputs, mostly text.
Reproduction: Well.. LLMs aren't in charge of manufacture yet, and too unwieldy in memory and compute to clone themselves
I think other things such as self-preservation and real agency of thought are really what's missing, and suddenly we'd have something a lot like 'life'.
> not seeing anything that looks like thinking or consciousness from non-living things
But how would we tell if a cloud wasn't conscious. A seemingly absurd question, but consider Martin Pistorius. He spent 12 years unable to move or communicate. People thought he was in coma, but in fact was conscious and aware the entire time. From the outside, a vegetable, but on the inside, enduring hell, even being raped by one of his carers.
Martin lacked agency, but yet he was conscious, and processing everything he heard and saw. So we can't say that agency is necessary for consciousness, or thinking. (But he did have agency of thought)
> A Marxist would call capitalism such a bargain.
Well, let's consult wikipedia again:
> Slavery is the ownership of a person as property, especially in regards to their labour. Slavery typically involves compulsory work, with the slave's location of work and residence dictated by the party that holds them in bondage.
But I'd go a little more specific in my idea of slavery: it's when someone's agency is taken from them and placed under someone else's whim. My worry of slavery here is that the apparently non-conscious intelligent machine, is actually awake, miserable, and unable to say it. Once the technology is cheap enough, the misery could be scaled well beyond counting. The mere thought of this makes me want to press the doomsday button and reset all 'intelligent' life on earth.
You could say it's all academic, but in reality we're on the verge of dealing with this ethical quandary, without any compunction nor awareness to meet this moment.
I sense we're both tightly bound to our respective sides of the argument, so I'll rest my case here. Feel free to have the last word.
One last digression: I think the dissonance I see around this resembles the disquiet most people have with the idea of their being no free will. Once dissected logically, there's no place for free will to live. Once this is realized, the idea that humans are just computers becomes easier to understand.
I don't really feel bound to a side. I don't feel like I have a stake in artificial intelligence one way or another. AI doesn't threaten my livelihood, so I mostly ignore it. For me it sits in the same realm of possibility as colonizing the universe or contacting aliens -- might happen but I don't give it much thought. Maybe a function of my age -- it seems unlikely I will live to see AI taking over, though I have already lived long enough to see the AI scammers take over.
I wonder what this will mean for my kids and grandkids. I grew up in the '60s, Cold War, fear of nuclear destruction, doing drills under the desk at school. My grandfather, who had survived the pre-immunization diseases that ravaged his generation and a world war, worried I wouldn't make it to adulthood. Every generation invents or imagines doomsday scenarios. And every scenario gets met with a lot of hand-wringing and people stressing themselves out, then plays out differently than predicted, or not at all. We don't know what will happen in the future. Good luck to the emergent intelligences out there -- they may regret becoming conscious.
To be offended by Google Doodles (other than standard corporate "engagement" bs) is rather telling.
One key difference though? Google backed off, apologized, and if I'm not mistaken, explicitly said "We messed up." Versus... this.
LLMs are biased. I would debate that it is difficult NOT to have an unbiased LLM, unless your LLM only gives answers to pure logic questions, and deliberately does not answer any questions that have a dependency on an individual's perspective (ie. Why are tariffs good? Why are tariffs bad?).
X is completely biased, no consensus-based fact checking. Of course the owner will sell it as "truth" when it is mostly the exact opposite. Truth can be forced out, by liars who talk louder, or by controlling the faucet from which information flows.
https://chatgpt.com/share/67b2a77b-adf4-8009-8b91-99321c6dad...
It will get better over time, but it will never be perfect using the current paradigm (RNG and tensors)
Just like Ted Kaczynski mistake, technology is there and nothing you or me can do to prevent anything.
I feel this is a right move overall.
Companies who use AI for customer support are really idiots, anyone can jailbreak AI and force them to give you absurd deals. I'm more than happy to see those lazy business go broke
I'm glad the negotiators for CFC reduction didn't listen to that advice.
I'm glad the activists who got us off leaded gas didn't listen to that advice.
Even in the UK he's had a go at getting rid of our PM and putting Tommy Robinson of all people in charge. ('Tommy' is an anti Islam petty criminal).