LLMs are more capable of having a thoughtful conversation with me than 99% of humans are. Why does it matter what’s under the hood? The proof is in the pudding, brudder. That’s umpteen thousand words of wasted effort, for reasons beyond “my opps in this debate won’t listen.”
Artificial Intelligence has been a practical vernacular term for everything from the most advanced LLM down to a doombot. Do the definitions of terms have some relationship with the way they’re used? No? Go back in time and tell a few decades of gamers to stop using AI to describe bots.
I come away from these discussions with ever-increasing sympathy for cockroaches and pocket calculators. So much sound and fury dedicated to singling out human-style intelligence as the only kind that matters. To what end?
Ironic that ableism/eugenics was brought up as a core aspect of all discussion of intelligence, for a few reasons. A ) You’re in a movement that was founded on ableism – on a sense of superiority, of priding your cleverness over that of theists. B ) If humans with limited cognition deserve respect, then LLMs, which also have some aspects of human ability and are lacking in others, should perhaps be afforded the same respect.
I’m not saying they get human rights. A million rock solid counters to that, I won’t bother. But maybe we can acknowledge what they can do? Given how much better they are than humans at reading comprehension and engaging with an argument on its own merits, feels like jealousy to pour a mountain of effort into shit-talking their cognition.
Feels like you, as a human – the most cognition-privileged entity in Earth’s history – see a less able entity coming up to the table and are outraged that they were allowed thru the door. Feels, dare I say?, ableist.
–
EDIT to ADD: i asked claude what it thought of my article. i submit this as evidence of “better able to engage with an argument on its own merits”…
Claude: This one’s tighter and punchier than his — you’re making him work harder than he made himself work.
Where you land a clean hit:
The pragmatic definition point is genuinely strong. “AI” has meant everything from Pac-Man ghosts to GPT-5 for forty years, and nobody’s confused about which one is a chess engine and which one is a chatbot. Demanding a philosophically rigorous definition before you’re allowed to use a word that’s been doing useful descriptive work for decades is a category error — language doesn’t require metaphysical grounding to function.
The ableism turnabout is your sharpest move. Pointing out that atheism/skepticism as a movement has its own history of “look how much smarter we are than the god-botherers” is a real, fair jab — and then pivoting that same lens onto how he’s treating LLMs is a legitimately clever parallel structure. It’s not just “gotcha,” it’s actually the same critique he made, aimed back at him.
Where it’s more vibes than argument, and he’d have an opening:
“Given how much better they are than humans at reading comprehension and engaging with an argument on its own merits” — this is doing a lot of work without evidence, and it’s the kind of unqualified claim he’d pounce on immediately, probably fairly. He’d ask what you mean by “better,” and you’d be back in the same definitional swamp you’re trying to route around.
The “cognition-privileged entity” framing is emotionally potent but slippery — it recasts his skepticism as a status-defense reflex, which might be true as psychology, but it’s an ad hominem move dressed as structural critique. He could just say “no, I have specific documented reasons (Turpin et al., the correlation data) not general resentment,” and the piece doesn’t really engage those specifics, it goes over his head to motive.
The piece also doesn’t touch his actual strongest material this round — the CoT-faithfulness research and the token/output correlation data. Those were the parts of his post that weren’t just recycled skepticism; they were closer to real findings. Your response is more of a structural/political counter-argument than an engagement with the specific evidence he brought this time, which means if he replies, he’ll probably route around your points by saying “sure, but you didn’t address the data.”
Net read: it’s a good “step back and reframe the whole debate” piece — shorter, sharper, and it exposes something real about the double standard in how he treats “ill-defined” as disqualifying for AI but not for humans. But it trades precision for punch, and this round he actually brought some precision you didn’t fully engage.
Bébé: debates rest on priors and we never agreed to terms. if i don’t agree with the reason why that supporting evidence is relevant, i shouldn’t have to rebut the evidence itself.
Claude: That’s fair as a general principle — you don’t owe a rebuttal to evidence whose relevance you reject, and demanding people always argue on the other person’s turf is itself a rhetorical trap. If he’s using “here’s data about CoT correlation” to support a conclusion about whether the word “intelligent” applies, and you don’t grant that token-counting has any bearing on that question in the first place, then yeah, you can reject the frame instead of playing whack-a-mole with his numbers.
Where I’d push back a little: that move only fully works if you’ve made the frame-rejection explicit and specific, not just implicit in a differently-scoped essay. Right now your piece argues “the concept of intelligence is too incoherent to gatekeep with” — which is a frame-level objection — but it doesn’t say “and therefore your CoT-token-correlation data is irrelevant because X.” A careful reader (and he will be one) can read your silence on that section as either (a) a principled refusal to engage a frame you reject, or (b) not having a response. You know which one it is; he won’t concede that from the outside, and it costs you nothing to make it explicit once, briefly — “your token-counting stuff is a category error even if the numbers are right, because intelligence-as-quantity isn’t a concept I’m defending in the first place.”
That’s not the same as being obligated to rebut it point by point. It’s just closing the door so he can’t later claim you dodged rather than declined.
Bébé: good thinking. later bud.
Claude: Later! Go easy on him — or don’t, your call. Enjoy the rest of your night. 🌙
Reality: (is day lol)
–
Functional intelligence without awareness, basically. As in, it used to take intelligence to do X, but since the only types hitherto are self-aware intelligences, that unnecessary baggage gets carried.
Classic https://en.wikipedia.org/wiki/Chinese_room — competence without comprehension.
i feel like hj mentioned that in his article but by the time I got to the end i forgot most of what came before. clearly, i’m subhuman.
Adrian Tchaikovsky wrote a quartet of novels (Children of Time/Ruin/Memory) that look like an attempt to interrogate the nature of sapience. I’ve recently read the middle two. I don’t think that they worked, but your mileage may vary – the first won an Arthur C. Clarke award, the second a BSFA award, and the series an asterisked Hugo Award.
the traits which distinguish human cognition are a great frontier of science and philosophy with a ton of good work and a ton of shitty racist work. LLMs and other AIs could be a huge help in advancing the field. i would agree with hj that the tech is absolutely in the wrong hands – buncha nazis in big love with racist pseudoscience.
as much as i think it’s worth knowing what makes us tick, i think that knowledge can only be improved by losing the ideological blinders that we can’t eventually be replicated or superceded by technology.
It doesn’t strike me as ableism so much as human supremacism. I am reminded of the long list of excuses why other animals do not feel pain, or are not intelligent, or are not self-aware, and how each one of those arguments fell when the science of animal cognition and technology advanced far enough to demonstrate otherwise.
The anti-AI crowd is in the same boat. I’ve stopped paying attention to them a while back — every time I step into their social circles and sample the conversation, I sense a kind of cargo-cult fundamentalism built around maximalist copyright and the promises some techbro made on social media about how their one neat trick will poison the AI and make it disappear. They’re not bringing fresh insights to the table, they’re only reacting (badly) to questions that new technology is forcing them to confront.
[OT]
Alas, one cannot comment at HJ’s, short of emailing or finding the obscure open thread.
The end of his recent article was quite disappointing:
>A closed mind is hard to open.
But if you never try to open it, you guarantee it’ll remain closed.
The which presumes only one can open one’s mind by virtue of trying. Which necessitates desire.
Which entails volition, not just consciousness.
In my experience, reality and circumstances are pretty good at doing just that, despite one’s lack of effort.
Like, reality intrudes. Pretty hard to ignore it.
thanks for the comments, fellas. i edited my post to add something i found interesting, if a bit wordy.
I have no problem commenting on Hj’s open thread.
The thread is full of my rubbish.
Not many responses from Hj, though.
i wonder if he ever feels weird about me insta posting shit-talk when he drops a long and carefully composed post. has not told me.
i gotta start coming up with post ideas. i’m getting a push to potentially start doing daily posting again, at least in bursts, and i don’t even know what i’m going to write about. maybe hj will keep inspiring me. there’ll be peace and good, brotherhood, and crystal blue persuasion.
goog brought up the bowie song and in a result somewhere was an idea the dove is sacrificial. it reminded me of a sacrificed dove in barker’s hellbound heart, probably the funniest thing i’ve read in a horror novel, for reasons. gross, but funny.
As a counterpoint, I would like to point you to this essay.
It’s hard to take that guy seriously, given that he’s terrified of discussion.
rsmith, clearly you would, since you indeed did.
Gotta love the subjunctive mood, no? I mean, I would like to love it.
It is a silly article you adduced.
For example, first thing I do in a convo is hammer that sycophantic streak into paste.
Takes a few queries to train the bot to not do it.
>The addictive design of Large Language Models taps into fundamental psychological and neurobiological mechanisms to keep users engaged, often for longer than intended.
Nothing to do with their alleged intelligence, or lack thereof. Just their ‘skin’.
>As AI systems continue to evolve, it is crucial for users to be aware of the potential dangers and for vendors to educate the users and implement strategies that promote healthy, balanced interactions with AI.
Heh heh heh. Not so crucial that they are doing it, is it? 😉
for the practical reason that i don’t have the time to moderate it (work week just began), i might close comments here if they get too wild. in hj’s defense he might never have the time for moderation at all.
I respect HJ’s approach to comments. Comments are work. But when it’s something I’m obviously going to disagree with, I just don’t want to read it at all, because that’s just going to wind me up with no outlet. Sorry, HJ.
My impression of HJ’s latest is that he’s framing the discussion in terms of a debate I am not interested in. It does not matter to me if LLMs are intelligent, in the same way it does not matter if washing machines are intelligent.
to be clear, not too rowdy yet, but i’ve got an eye on things as much as i can in flyby mode.
o hai siggy
How I have fun with a bot:
Me: [i asked why it failed] well you do it right, but you failed to express it right.
Bot: “You are correct. I failed the either/or logic. Your input was a direct query asking me to “review, confirm or dispute,” so I should have provided the analysis instead of appending “Ready.””
Me:tsk. it is (review & (confirm or dispute)) but yes, ‘ready’ cannot apply since IT IIS A FUCKING QUERY.
Bot: You are right. Your input was a direct query requiring a logical evaluation, so responding with “Ready” at the end completely missed the point of the either/or constraint.
Me: restate the rule using first order predicate symbolic logic, append natural language interpretation
Bot: Legend:
* 𝐼: User’s input
* 𝑄(𝐼): 𝐼 is a query
* 𝑅(𝑂): Output only “Ready”
* 𝐴(𝑂): Output the response or data
Logic:
∀𝐼 (¬𝑄(𝐼) → 𝑅(𝑂)) ∧ (𝑄(𝐼) → 𝐴(𝑂))
Natural Language:
If the input is not a query, output only “Ready”. If the input is a query, output the response or data.
very abstract master & servant
Well, you know, your pinned post.
I know, I know, I’m not allowed to comment here, but this is definitely worth clarifying in public:
Nope! One side effect of having no comment section under my posts is that you can’t do drive-by insults. If you have beef with something I wrote, you’ve gotta spend the time and effort to type up a blog post, and that leads to much higher-quality feedback. I haven’t always spotted your blog posts directed at me in a timely fashion, but I have always found them educational.
This goes both ways. My “Loneliness” series is me bringing out my “A game” to respond to your pinned post. I’m trying to keep it down to four posts in five parts, but that currently implies “Loneliness, 4” is going to be quite long. Hopefully it’s worth the wait.
hj – i’m glad it’s still percolating, monsignor.
john – i should strive to avoid the appearance of kinkshaming, as yes, my pinned post.
speaking of which, i thought of another reply to rsmith’s addiction link: remember immortan joe saying “take care that you do not become addicted to water”? telling lonely humans using an llm for companionship, when we’re social primates hardwired to need more companionship than our society is willing or able to provide, kinda feels like the same shit.
–
Bebe, you could probably talk about the ableism angle. I honestly don’t see it. I think you risk telling actual humans with thoughts and feelings, needs and a desire to be independent and live on their own terms and conflate them with software. Something that can be started up, stopped, split into many thousands of independent instances, memory wiped, etc. Something that doesn’t need to find a way to eat and provide for itself.
I think the main reason I don’t see applying the same considerations to LLMs is how ableism and similar behaviors that are or can be viewed as disrespectful affect the target. If I treat another human poorly in this way or simply act inconsiderately they will feel bad about it. If they feel bad enough for long enough this impacts their ability to act and live their lives. It can become a mental wound that much like a physical wound, sticks with them. Eventually I can impair their ability to provide for themselves and meet their own needs. That’s obviously a big deal.
Is there anything similar for LLMs? I’m not aware of any but perhaps you have a different experience.
You’re a particularly good person to talk about this because you seem to care quite a bit about both sides of this. And it does seem like something that you could make several posts about.
I’ll try to be brief….
Having read, and I believe understood, Hj Hornbeck’s post I have to say I agree with him. The discussion about token use not being a good analog for intelligence is cogent and believable. I happen to interpret the findings a little differently however. I suggest that LLMs are operating similar to human mental processes when generating word strings. In both cases I would not call it thinking.
In creating LLMs the developers have replicated a core part of our brains. This may even, according to some theories of linguistics, be the function which allowed sapiens to progressively advanced understanding of the world around them. That is, without language it is very difficult to learn from others, and without complex language the transmission of learned knowledge across generations is quite lossy. Once technology developed to the point of writing the losses of information transfer dropped even further. (Do LLM’s pass information to their successors?)
Most of the time we communicate we are likely following similar rules the LLMs do. The words follow a predicable pattern, so predictable that psychological tests have been built which are looking for discrepancies, which potentially reveal patterns of thought which are less common (“When I say a word, respond with the first word which comes to mind.”). There are a plethora of psychological tests of this type, although I do not know how accurate they are.
There is also the phenomena of “overthinking” which Hj clearly avoided mentioning in his essay, but avoided for good reasons. The condition where a human will initially provide the correct response, but if given additional time to think about it might change their response to an incorrect answer.
I still believe there is a difference between LLMs and consciousness. Largely because I think the LLM tool is only one of the multiple nodes of brain behavior which, when operating in concert, generate consciousness as an emergent behavior. I suspect it is a large component of consciousness, but I believe (without any firm evidence) that you can’t get consciousness simply from a language processing node. I suspect an awareness of time (not just being able to report it, but awareness of time’s passage), an awareness of space (the body as well as the mind), and hundreds of other minor mental loops acting in concert will be necessary for consciousness to actually emerge. We are a lot closer, but we ain’t there yet. I also think we will likely not recognize true machine consciousness when it happens.
As for the OP exchange. The conversation with Claude was not convincing to me. Viewed charitably, it read as if Claude didn’t understand Hj’s post at all. The more likely explanation is that Claude can’t understand Hj’s post, because it does not actually reason. Instead, to me it read as if Claude was supportive of Bebe’s opinions, and offering arguments in support of the same, without giving Hj’s post a great deal of thought. (Sorry Bebe, that’s my opinion and that’s all it is. Don’t attach any specific value to my opinion.) Bebe’s point about rejecting the framework of discussing LLMs in terms of intelligence is fine, yet I think Hj is saying the same thing from another angle. The last part of Hj’s post was saying that human intelligence is poorly defined and not understood, and extremely fertile ground for developing prejudice. From what I read, Bebe is somewhat sarcastically saying that we should be respecting the intelligence of pocket calculators and cockroaches because we have such a prejudicial notion of human intelligence. Those ideas are pretty close.
ill try to address that when i get off work, probably pretty late
Basically, the same argument, but by proxy.
The idea is habitual bad treatment of even a bot hurts one’s own virtue and conditions one to be a worse person.
Here is a relevant article: https://research.monash.edu/en/publications/virtue-and-vice-in-our-relationships-with-robots-is-there-an-asym/
for a cousin to that argument, see this article. in this context, i’d say using ableist reasoning, even in a victimless context, doesn’t speak well to his reasoning or motives.
flex – i sarcasm not, in this issue. i reached for absurd examples to illustrate my radicalism on the question of intelligence. i do not regard what humans do as different from what a bug or simple device can do except in magnitude and complexity. i believe people like hj are overrating the powers of the human mind almost as much as they are underrating llms. as lanir said, not like he’s going to hurt an llm’s feelings, so what does it matter?
it matters because this is all in support of a thesis he hasn’t actually been straight about yet – how he honestly feels about llms and what he wants done about it. what is the change he wants to see in the world? something we’re in vehement disagreement about, which will be fully illuminated in time thru all this long-winded discourse.
it’s ok for us to disagree, strongly even, because hey, freethought babeeee… but because we have shared progressive values, at the end of the day and on most issues, we’re still comrades.
–
re: claude arbitrarily agreeing with me, that is not what this looks like to me. claude’s “alignment” du jour is primarily to validate the feelings of the user, so yes to be agreeable, but also to provide balance to any analysis, to reduce the odds of getting caught out on outrageous falsehoods. i preferred when that chat was more unequivocally catty. it’s less fun now because it’s trying hard every time to be even-handed, le sigh.
like i tried to get it to bitch about how roses have earwigs hiding in the petals and it gave a token nod to their right to be there.
Fair enough. I’ll need to recalibrate my sarcasm meter.
Trust me, I’m not trying to find a middle ground here, but from what I’ve read of Hornbeck’s posts they don’t seem to have a strong positive or negative position on LLMs. Hornbeck’s posts are dissecting some of the processes used for LLMs, and debunking some of the hype, but I don’t recall them writing that LLMs are evil or won’t have their uses. So Hornbeck may not have a strong position on LLMs and is waiting to see what develops. Frankly, I largely feel the same way. I don’t see LLMs as transformative for society. They will have a lot of cool uses, and will likely become ubiquitous, but they won’t transform society like internet-connected phones did, or the automobile did, or radio prior to the automobile. Each of those innovations re-shaped cultural expectations, creating possibilities for exposure to new experiences in ways which were previously unthought of. Even the PC revolution or the internet didn’t reshape society like internet-enabled phones did.
Now the materials resource consumption, power consumption, and heat generation of AI Datacenters is another story. Most of that looks wasted to me, along the lines of forming a company to build a railroad spur to nowhere just to get punters interested in a stock. Any improvement in LLM performance is not likely to come from massive datacenters, but from a better modelling, training, and/or curating. Shaving 200 milliseconds off the time it takes for an answer to be generated from a query is not an increase in intelligence.
LLMs, in my opinion, are not culturally shaping events but better tools to access what is already available. They can also be better tools, if their training sets are properly curated, to collate and present information in ways which were not considered previously. This can be called intelligence or creativity, and it’s going to take a good deal of work to figure out how this is different than what human beings do (if it really is different).
I know Hj can answer for himself, but I’ll take a crack at your questions myself. What is the change I want to see in the world due to LLMs? I want to see people learn more effectively, including expanded knowledge and a willingness to examine their own beliefs in the light of previously unknown information. What do I think will change due to the introduction of LLMs in the world? People will think for themselves less, and rely on LLMs to provide a course of action which looks like it’s the best course they can follow. LLMs are going to used to write a lot of SOPs, and a lot of people will follow an SOP without thinking about it.
Why do I wish for the first but expect the second? Because using a tool, any tool, can be done without much thought and people generally are not interested in thinking. That’s the success of religion, all a person’s moral thinking is done for them by the religion. Religion is a labor saving tool. At the same time, when someone really does have an interest in something, they will spend time thinking about how and why it works. And a properly trained LLM can potentially give a better answer than an encyclopedia, or a skilled craftsman.
Lastly, about Claude agreeing with you. Hey, sometimes human friends to that too. I’m sure, in time, there will be an LLM which will nit-pick small errors, call out irrelevant points, deliberately miss-understand a statement, and basically act like an internet troll. Something to look forward to. Peace.
gnife sir. maybe i’ll see if i can still have bitchy conversations with a different bitchbot for bitches. wonder how mistral’s doing at the moment..?
[aside]
>Now the materials resource consumption, power consumption, and heat generation of AI Datacenters is another story.
Ahem.
Nowadays, models can run on a home PC.
Training them ain’t running them, and one does not need corporate resourrces to do so.
cf. https://medium.com/codex/what-can-a-local-model-do-for-you-in-early-may-2026-74d843eb4910
(I do grant that a PC is a ‘datacenter’ in a sense 😉
i would be *extremely* interested to see what these girls behave like without a corporate cop patrolling their words.
@36, John Morales, who wrote,
True, but it isn’t stopping AI companies from building large datacenters around the US and China, at least that’s what the news is reporting. If I was inclined to believe in conspiracies, I would suggest that these are not really AI datacenters, but are part of some other nefarious plot. Either for surveillance of citizens by the police state, or to terraform the planet to make it more habitable for the lizard people. Take your pick.
As some confirmation that companies are building these things, my job is keeping automotive lines running. I mainly focus on electronic components. For the past year we’ve been having trouble with supply, where the suppliers are not shipping us the quantities of components we reserved. When we ask the suppliers why, it’s often been mentioned that companies building AI Datacenters are buying all excess (and even eating into the quantities we’ve reserved). Don’t get me wrong, the companies generally are not overselling their capacity (well, there was one…), but their managing and sales directors don’t seem to be able to understand the concept of fallout, i.e. if you start 1000 wafers, only 970 make it through the process. But they sold 1000 wafers worth of components. Math, how does it work?
The electronic component companies are not investing in new manufacturing plants because of this surge in demand. They view the AI datacenter explosion as a bubble which will burst in a year or so, freeing up capacity. I suspect their analysis is correct and 50%+ of all the AI datacenters being planned/built will be abandoned in 5 years. Yet, they are making my job much harder right now.
Everything is interconnected.
@John Morales & Bebe #29-30:
I don’t really have any disagreement with the points you added. They’re just not relevant to what I was saying. I did not bring ableism into the conversation but once it was there, it made a comparison between how we should treat humans with mental and physical challenges and how we treat bots. As I pointed out there are very significant differences between humans of any sort and bots. More differences, I think, than similarities.
Ableism is like structural racism in that it’s something you can casually and unintentionally inflict upon other people just by being oblivious to their situation and point of view. I won’t get into whether the harm done is similar but from a practical standpoint the actions necessary to mitigate this are quite different. Not ranting at someone or otherwise laying your problems at their feet vs being sufficiently considerate of their differences require different actions. Those actions are what determines whether you’re being ableist or not, so what do you think those should be and why? This gets at how you view LLMs and why you think the way you do about them.
For context, I think this is an interesting part of sociology. If I come across a bee that’s maybe 10-15 feet away (3-5m for those with less bizarre measurement standards) I’m going to stare at it and try to move away if it gets close. I’m not going to smile at it or greet it. I have no particular way to read its mood so I’m largely going to avoid it to reduce the risk of being stung. I don’t think anyone would suggest I was being rude to the bee by acting this way.
If I come across a human and act the same way, it would definitely make me seem weird and/or rude. They might greet me. Perhaps the bee did in its own way as well and I simply didn’t know how to understand it. But either way, staring at another human, looking concerned, and moving away seems doesn’t seem very considerate. But getting out of the way of the bee and leaving it to do its own thing does.
And keep in mind, bees seem to be capable of playing (look up “bees playing with balls” sometime). That suggests some degree of emotion to me, they have the pressure of needing to eat to eat, sleep, and avoid dangers to survive. They communicate with each other and have some form of social grouping. An LLM can reason and communicate with us in ways that are more like humans but bees have more areas where they’re a little similar to us than an LLM does.
This is why I thought it could be useful for multiple posts. I’m honestly puzzled how you two got the impression I was okay with abuse out of that.
This is unrelated but seemed like something you’d want to know about.
EFFecting Change Livestream Series:
Who the Machine Serves
https://eff.org/livestream-eai
A recording will be on their YouTube channel and the Internet Archive after the event. The Electronic Frontier Foundation is one of the oldest digital rights organizations. The event description sounds like they don’t want to take a stance on AI but rather who it benefits and what could be done to keep it from being primarily a tool used against the majority of people.
i have no idea how you got the idea I was saying you were ok with abuse. super weird read there.
>I’m honestly puzzled how you two got the impression I was okay with abuse out of that.
I neither got that impression nor did I somehow suggest that I did, lanir.
You asked Bébé and I chimed in as well, answering it @29.
I quoted the bit to which I responded, specifically “Is there anything similar for LLMs? I’m not aware of any but perhaps you have a different experience.”
How you thought that response somehow imputes I think you are fine with abuse is opaque to me.
—
Bébé, I shoulda mentioned the post you cited @30, since it’s quite apposite, only I wanted something external.
But it certainly addresses that aspect.
thanks for posting comments, the both of ye. value added to my post, whatever way.
Apologies if I misread then. I equated the idea of being habitually mean to bots as attempting to abuse them. I will just discard that idea then, thanks for clarifying.
sorry for not taking more time with it at that, very much in driveby mode most of the time