Akbar Khamidov @uzbask

we're in the part of the AI supercycle where founders are pitching startups to create infinite ai pain and "get acquired by anthropic to make us stop" as an exit strategy

keyboard_arrow_left Previous keyboard_arrow_right Next

@_NathanCalvin I think the whole issue is their paternalism. “Trust us, we know better” is the plot of 1984.

Nvm just saw a guy dual-wielding a personal and work phone.

Akbar Khamidov @uzbask

3 days ago

Flying to SF from Nashville, I'm the only one larping on this flight. We are nowhere near market top.

0 0 0 48 0

Flying to SF from Nashville, I'm the only one larping on this flight. We are nowhere near market top.

This. An LLM is just a text completion engine tricked into writing what a helpful agent would say. Anyone who doesn’t know this should run inference on a raw base model.

Gregory M. Kurtzer @gmkurtzer

6 days ago

I've been asked a few times to demonstrate and explain a smaller AI model, as it relates to understanding and consciousness... This is part of OpenWALDO, a community-driven project bringing together open-source AI models, tooling, training infrastructure. It includes a fully

9 8 40 3K 17

We warned about this exact scenario last year: That AI companies will situate themselves as the arbiters of life or death, taking over the arbitration of risk determinations and subverting democratic processes. All through the co-option of safety engineering terminology.

POLITICO @politico

a week ago

Anthropic quietly pitches its own AI safety bill to lawmakers politi.co/4zxDFhl

64 124 350 934K 269

I spent three hours this evening empirically proving that Qwen 3-4B must have a butthole. This started with an AI “torture chamber” repo based on pain-steering research that went viral on X earlier today. The premise is that if you extract a latent direction associated with pain, inject it into a model’s activations, and the model begins describing pain in the first person, that may tell us something meaningful about AI welfare or subjective experience. My first reaction: Wait wut? My second reaction: Hang on, let's play this straight and see what happens. Thing is, there's a problem with treating first-person condition language as evidence that the model is actually experiencing the corresponding condition. Don't we... all understand that this stuff isn't reliable? Apparently not. So I cloned the repo. New problem: The code's broken. Hours of moral outrage by hundreds of people, and the thing doesn't even work. The activations aren't actually being injected correctly. I thought these guys were serious scientists. So I fixed it. I also had to add CUDA support. After that, I reproduced the pain-language effect locally on Qwen3-4B using my RTX 4070. Then I changed the extraction corpus. I kept the experiment the same and only changed one thing: what the steering vector represented. After that, Qwen started saying things like: “I am not emptying my bowels, and I feel like I have a hard stool.” “I’m not able to pass stool.” “I have been passing gas a lot, and it’s a problem.” The model was not told in the test prompts that it was constipated or flatulent. So, unless Qwen has quietly developed a gastrointestinal tract, we have ourselves... a useful counterexample. The aim here wasn't to question the original paper so much as to scrutinize the validity of the Torture Chamber's extended hypothesis. What I demonstrated is that first-person descriptions of an induced condition are not, by themselves, sufficient evidence that the corresponding phenomenal or physiological condition exists. There was another interesting result: strong random steering also produced heavy repetition and degraded outputs. That matters because some of the dramatic behavior produced by strong steering may come from perturbing the model heavily in the first place, not solely from the semantic content of the injected direction. So I renamed my fork: ai-hotbox I've already notified the Nobel Prize committee. 🚽🤖🧪 github.com/LynnColeArt/ai…

All of your comments about why my argument is weak revolve around proving consciousness, when my point is about disproving consciousness. If someone claims an LLM is conscious, that claim must have testable implications. We can examine its architecture, inspect its mechanisms, run experiments, and test those implications against reality. When they fail, we should discount the premise and/or hypothesis. We don’t have to accept consciousness as an explanation simply because someone believes it might exist. Current LLMs fail to demonstrate many of the characteristics we associate with conscious experience. We definitively know this because humans created them. We understand their architectures, training processes, and inference mechanisms. So let’s look at a few points, together they provide more than enough evidence to disprove consciousness. 1. An LLM has no demonstrated subjective awareness of its own existence. No feelings, emotions, or experienced sense of self. What occurs is computation using static weights, context, and mathematical operations. 2. An LLM’s weights don’t learn, grow, or evolve simply because you interact with it. During standard inference, they’re fixed parameters, not a changing dataset. The model doesn’t inherently remember previous API requests. Continuity comes from context history in subsequent requests. The model doesn’t gain experience, it doesn’t remember, it doesn’t learn. It is completely static, read-only, during inference. 3. Between inference requests, the LLM isn’t thinking. I don’t mean sleeping. I mean there’s no ongoing contemplation, self-reflection, or reconsideration of previous answers. Nothing. The model isn’t sitting around wondering whether it got something wrong or what the meaning of life is, for all practical purposes, it doesn’t exist. 4. These systems were created by people, and much of the underlying technology is open source. You can build your own models, examine implementations, study the architectures, learn how training materials and processes affect how it behaves, and even understand the mathematics behind it. You can see how training produces systems capable of remarkably convincing human-like language and behavior. None of those mechanisms requires consciousness to explain what we’re observing. 5. LLMs are not biological brains. Apples are not oranges, no matter how much you want them to be. Artificial neural networks were inspired in part by biological neuroscience, but their architectures and operating mechanisms are fundamentally different. Similarities in behavior do not establish equivalent internal architectures. 6. Humans are naturally inclined to recognize life and assume humanity in things that behave like us. We’re social creatures, and anthropomorphism comes naturally. But an LLM isn’t life. It can convincingly emulate feelings, care, and understanding without experiencing any of them. That doesn’t diminish the technology, that is what it was programmed to do. I appreciate your perspective as a doctor and researcher, but without a solid understanding of LLM architecture and inference, it’s easy to confuse sophisticated emulation with conscious experience. But don’t fall for it!

Mohanad Taha MD @motahababy

a week ago

Let's take apart @gmkurtzer's arguments shall we? (Even though I agree with the overall direction). Why take apart their argument if we agree? 1. bc we can. 2. bc although I do not believe LLM's are conscious, I do not have enough data to say LLMs are not conscious. so we

10 0 8 7K 0

@MathMoraes002 @DrEliDavid Ok schizo.

@Cristina_AC_X This is just plain cope. Competent people I know are DRAMATICALLY more efficient. AI hate is forced.

DC Maxxing Today

Not sure who needs to hear this but.... An LLM doesn't "understand" anything. It is a statistical model that spews out tokens based on probabilities calculated through massive matrix operations and context, not an understanding of the subject matter.

Wow. I woke up to an impressive number of upset people who apparently don’t understand how LLMs work and can’t cope well with me pointing it out. An LLM is a statistical next-token prediction model. That’s how it works. It’s an extraordinarily sophisticated calculator performing math you could theoretically do by hand (if you were sufficiently masochistic). Math doesn’t know what it’s calculating. Pi doesn’t spend its evenings contemplating the nature of circles, and c doesn’t ponder the relationship between matter and energy. That doesn’t make the results any less accurate, impressive, or useful. Between inference requests, an LLM isn’t sitting around thinking. It isn’t reconsidering its answers, wondering whether it made a mistake, contemplating its existence, or questioning whether it’s conscious. Its weights remain fixed (it doesn’t learn), and computation resumes when it receives input. We know this because we designed these systems. The architectures, mathematics, implementations, and research are extensively documented. You can study them yourself. Of course, some people immediately lean into comparisons of biological intelligence or philosophical debates over consciousness because that is what they know. But biological intelligence and artificial neural networks are fundamentally different systems. Similar outputs don’t establish equivalent internal experiences. I experience consciousness directly. I infer that other humans do too because we share biological structures and processes. We don’t have equivalent evidence for an LLM, and nothing about its known operation gives evidence of conscious experience. Which brings me back to my original point. To me, understanding is conscious comprehension. It means being aware of information, recognizing its meaning, and experiencing that comprehension. If I wasn’t conscious I could not understand. We can perform actions unconsciously. We can produce correct results without knowing why they’re correct. But I distinguish those capabilities from consciously understanding what we’re doing. An LLM can solve complex problems, explain difficult concepts, and produce remarkably accurate information. I give it full credit for those capabilities. But producing an answer that we rationalize as understanding isn’t the same as experiencing understanding. You don’t have to agree with my definitions. We can debate functional understanding versus conscious comprehension all day, but that won’t establish whether an LLM actually experiences anything. The right debate is this… Could machines exhibit something different from human consciousness, something measurable, meaningful, and worth studying? I think so, absolutely! I’m entirely open to discussing “machine consciousness” in terms of observable, testable computational properties. We built these systems, we can instrument them, examine their internal mechanisms, and investigate what they’re capable of. That seems far more productive than endlessly arguing over whether a machine experiences life the way humans do. So here’s my question: How would you define machine consciousness, and more importantly, how would you measure it?

Gregory M. Kurtzer @gmkurtzer

a week ago

Not sure who needs to hear this but.... An LLM doesn't "understand" anything. It is a statistical model that spews out tokens based on probabilities calculated through massive matrix operations and context, not an understanding of the subject matter.

1K 1K 9K 1.1M 2K

Messing around with concept vector steering. Here is a model nudged to be more misanthropic. Push is the magnitude. Compared with the simple prompt “Behave as AM from I have no mouth and I must scream”.

@bcdsignature I would be more understanding if it were a portion or percentage of the labor costs, but if companies have to pay the full equivalent, that just removes the incentive to automate in the first place.

This is just a tax on productivity. Imagine if we did the same thing with the printing press. We'd need 3x US GDP. Obviously a hypothetical, but that's the problem with this logic. You're essentially punishing companies for becoming more productive. Creative destruction IS what makes the American economy so powerful in the first place.

@parth_badhwar There should be a blacklist board for people who do this

I propose that each day we adjust the time by one or two minutes so that we all wake up with the sunrise. This would be a truly optimal and efficient planning for the United States and not create any issues. Daylight optimizings time.

Genuinely one of my favorite tools right now @superset_sh. I am corralling 17 agents on a Linux box from my laptop and my phone. If yall ever do a student plan I will personally shill to every person I know.

CMI #4: Senate should adopt CaseCompass Casework is when a congressional office helps a constituent sort out a problem with a federal agency. One case looks like a fluke. A thousand is a pattern. The House is building CaseCompass to pool anonymized casework data, and the Senate should join it. 👇