LLMs aren't sentient; they lack the physiological states required for sensations like hunger and pain, and thus can't have subjective experiences of such states
Artificial general intelligence (AGI) is the term used to describe an artificial agent that is at least as intelligent as a human … Bluesky: @motomatters.bsky.social and @zitron.bsky.social X: @vsikka , @eshear , @landay , @mattrosoff , @oshanjarow , @dioscuri , @idavidrein , @ericlandgrebe , @jonasmtzgr , @hamandcheese , @drfeifei , @dioscuri , @sean_a_mcclure , @eshear , @psychosort , @ylecun , @mealreplacer , and @chrszegedy . LinkedIn: Ross Dawson Bluesky: David Emmett / @motomatters.bsky.social : This is a fantastic piece on why Large Language Models are not in any way, shape, or form, sentient. And this is also why they are the wrong direction if you were really interested in building actual Artificial General Intelligence [embedded post] Ed Zitron / @zitron.bsky.social : Yeah there's no way that generative AI gets us to approaching AGI. They are not building it and it is not what's coming out of ChatGPT or anything related to ChatGPT [embedded post] X: Vishal Sikka / @vsikka : A wonderful writeup in @time by John Etchemendy, a great teacher of my life, and my friend @drfeifei both co-directors of @StanfordHAI on the non-sentience of LLMs. https://time.com/... Emmett Shear / @eshear : In colloquial language of course this might be seen as the same thing, but if you expect to write about these things as an intellectual you need to be more precise or I just can't take you seriously. The article doesn't even start to grapple with the real questions. James Landay / @landay : Yes, LLMs are not sentient. Amazing that people need this explained to them... Matt Rosoff / @mattrosoff : Nobody thinks they are. @oshanjarow : As the article briefly alludes to — asking whether a particular version of AI is conscious distracts from the real question. The more interesting — & actually unsettled/ debated by relevant experts — question is whether consciousness is possible in non-biological substrates. [image] Henry Shevlin / @dioscuri : I disagree with claims that we *know* LLMs aren't conscious. Fairly unlikely, perhaps, but not inconceivable that the functional dynamics constitutive of subjectivity could be realised in large Transformers. Not enough scientific consensus about consciousness to rule it out. David Rein / @idavidrein : @drfeifei ... there are so, so many things wrong with this article. “One of the essential characteristics of general intelligence is ‘sentience’” False! We're confident that animals without human reasoning capabilities are sentient, and it's perfectly coherent to consider intelligent Eric Landgrebe / @ericlandgrebe : I like this view a lot. I think it's also relevant for more broad questions in alignment around where the agency in ML models is. It especially becomes tricky due to conditional generation, where the behavior depends heavily on the prompt itself. Jonas Metzger / @jonasmtzgr : @drfeifei ... your argument boils down to “AI doesn't have a body and hallucinates statements about experiences it cannot possibly have, therefore it can't be sentient”. But how do we know these are dealbreakers? You say yourself that we don't know how sentience emerges. Except in one way, Samuel Hammond / @hamandcheese : This is a bad argument imo. Humans have physiological signals of pain, hunger, etc., but the conscious states are all processes in the brain. Hence phantom limb syndrome, dreaming, or the ability to induce tastes and smells from neural stimulation. The truth is that we simply [image] Fei-Fei Li / @drfeifei : Is AI sentient? My friend and colleague Prof. John Etchemendy, a renowned professor and co-Director of @StanfordHAI , just co-authored this piece to debunk the claim that today's LLMs are sentient @TIME https://time.com/... Henry Shevlin / @dioscuri : @mealreplacer “All sensations... are the result of physiological states that an LLM simply doesn't have. Consequently we know that an LLM cannot have subjective experiences of those states.” So I guess someone who's had an arm amputated can't possibly feel any kind of ‘phantom pains’ there 😉 Sean McClure / @sean_a_mcclure : @drfeifei ... His argument rests on the assumption that physiological states like hunger can only be embodied in the specific manner humans embody them. He should be asking if an AI can be hungry for *resources* not food. And the “only predicts the next symbol” argument has always been Emmett Shear / @eshear : LLMs are simulators. They might contain sentient beings or not, but the LLM itself is no more sentient than the laws of physics are. If there's sentience, it's existing a layer up. Brian Chau / @psychosort : I will bite the bullet. The Bible, Hamlet, and Oyasumi Punpun are definitely sentient Yann LeCun / @ylecun : @ChrSzegedy No they don't. They simply don't have the required architecture for it. Julian / @mealreplacer : Though I highly doubt current LLMs are capable of having subjective experiences, it's an important and poorly understood question — one that deserves better than a “debunking” by an op-ed that doesn't even really attempt to engage with different theories of consciousness. Christian Szegedy / @chrszegedy : I think the arguments in this article only support that LLMs don't have a human-like sentience. It does not rule out that they have some different kind: maybe LLMs feel some “satistfaction” when it produces the sentence “I feel terrible”, as it enjoys predicting text. LinkedIn: Ross Dawson : The debates on whether AI is or will be sentient have started and will never end. Here Stanford luminary Fei-Fei Li argues that AI isn't sentient as it doesn't have subjective experiences. …
Context & Ripple Effects
This intervention joins a recurring dispute over whether convincing language behavior warrants claims of consciousness, following earlier debates over researchers’ claims that AI could be sentient. It argues that AGI discussions should not collapse intelligence, language fluency, and subjective experience into one category.
It also aligns with work highlighting the limits of what LLM outputs establish about understanding, including Emily M. Bender’s work on what language models can and cannot do. The immediate significance is conceptual: it supplies a physiological criterion for challenging sentience claims about current systems.
First-order effects
- The article gives researchers, product teams, and commentators a specific basis for describing current LLMs as non-sentient: they do not possess the bodily states the argument treats as necessary for sensations such as hunger or pain.
- It sharpens the distinction between AGI as a capability target and consciousness as a separate, unresolved question, even as some researchers dispute whether biology is required.
Second-order effects
- Developers of conversational and companion-style AI face more pressure to distinguish humanlike interaction from evidence of inner experience, especially when models speak as if they feel or suffer.
- The disagreement shifts evaluation toward what evidence would count for non-biological consciousness, rather than treating fluent self-description as sufficient proof.
Third-order effects
- If this framing becomes influential, AI discourse may increasingly separate performance benchmarks from claims about minds, making anthropomorphic product design a governance issue rather than merely a branding choice.
- The unresolved dispute over whether consciousness requires physiology could become a lasting fault line in AI ethics: one side will demand embodied evidence, while another may seek substrate-neutral tests.
The trend: As LLMs become more socially convincing, the AI sector is moving toward stricter distinctions between simulated affect, demonstrated capability, and claims of consciousness.