Hacker News Reader

Comment by antirez | Hacker News Reader

LLMs keep showing more and more they are the wonder of AI that we awaited for decades: talking machines that every two months do progresses that two months before were anticipated impossible because of <put here some limit that was actually in the prejudice of the skeptical AI community> (just stochastic parrots, no reasoning possible without symbolic representations, there are no longer tokens, ...)

At the same time, part of the scientific community continues to diminish what was accomplished and the steps that are being made. A few months ago LeCun arrived to tell new researchers to move away from LLMs since they are a dead end: imagine the disservice he made to the surely non-zero folks that followed the advice, putting themselves out of the AI research that matters. (Incidentally, this skepticism of the Meta AI head must have something to do with the fact that Meta, despite the huge efforts allocated, produced the worst LLM among Anthropic, OpenAI, DeepSeek -- I bet Zuckerberg is asking questions lately).

It's very hard to explain this behavior if not by psychological denial.

[EDIT: you can't see the score of this comment, but I can: it's incredible how it goes from 3, to -2, to 1, and so forth. The community is split in two, and it is pretty sad since this is not a matter of taste or political inclination: there must be a single truth]

View on HN

I get the sentiment, but I actually think some skepticism in the system is healthy.

Billions are flowing towards LLMS, and Sam Altman will overpromise AGI is just around the corner and the days of jobs are gone to fill his coffers to anyone that will listen.

Additionally if we begin to use these things in real production environments where mistakes matters, knowing the exact limitations is key.

None of this takes away from the fact that these are exciting times.

Yes there is another part of the community that overhypes everything. But I can expect that from a CEO of an AI company (especially if he is Altman), but from researches? Also the fact that LLMs may reach superhuman expertise in certain fields in a short timeframe (a few years), since reinforcement learning is starting to be applied to LLMs may no longer be a totally crazy position. If it is possible to extend considerably the same approach seen in R1-Zero there could be low hanging fruits around the corner.

This article is about things which aren't limitations anymore!

You are applauding it as pushback for pushback's sake, but it's an article about limitations in biplane construction, published after we'd already landed on the moon.

Is there any evidence that these fundamental issues with compositionality have been resolved or are you just asserting it? Has the paper been replicated with a CoT model and had a positive result?

Well, yes — because modern models can solve all the examples in the article. The theory of compositionality is still an issue, but the evidence for it recedes.

I think most of the issue comes from the challenge of informational coherence. Once incoherence enters the context, the intelligence drops massively. You can have a lot of context and LLMs can maintain coherence— but not if the context itself is incoherent.

And, informationally, it is just a matter of time before a little incoherence gets into a thread.

This is why agents have so much potential—being able to separate out separate threads of thought in different context windows reduces the likelihood of incoherence emerging (vs one long thread).

Actually, maybe “cybernetic ecologies” are closer to what I mean than “agents.” See Anthropic’s “Building Effective Agents.” https://www.anthropic.com/research/building-effective-agents

>I think most of the issue comes from the challenge of informational coherence. Once incoherence enters the context, the intelligence drops massively. You can have a lot of context and LLMs can maintain coherence— but not if the context itself is incoherent.

As a non-expert, part of my definition of intelligence is that the system can detect incoherence, a.k.a reject bullshit. LLMs today can't do that and will happily emit bullshit in response.

Maybe the "gates" in the "workflows" discussed in the Anthropic article are a practical solution to that. But that still just seems like inserting human intelligence into the system for a specific engineering domain; not a general solution.

I can’t communicate enough how the skepticism (“this is just hype” or “LLMs are stochastic parrots”) is the vastly dominant thought paradigm in European academic circles.

So instead of everyone having some enthusiasm and some skepticism, you get a bifurcation where whole classes of people act as the skeptics and others as the enthusiasts. I view the strong skeptics as more “in the wrong” because they often don’t use LLMs much. If you are an actual enthusiastic user, you simply can’t get good performance without a very strong dose of skepticism towards everything LLMs output.

> I can’t communicate enough how the skepticism (“this is just hype” or “LLMs are stochastic parrots”) is the vastly dominant thought paradigm in European academic circles.

I'm very curious. If you don't mind taking the time to elaborate, will you give a few examples of such skepticism/naysaying? Thank you.

Some "academia" types I meet seem to be struck in this "skeptic vs enthusiast" discussion.

How do we harness LLMs’ potential while rigorously mitigating harm?

I don't think everyone shares those doubts. The first time you catch an LLM in a lie is sobering, but there are lots of areas, and thus lots of users, for whom it doesn't hallucinate for, because they're asking softball questions and it doesn't end up hallucinating, or hallucinations just really aren't aren't that big a deal. (eg an LLM horoscope generator or using it write sci fi.)

so while we're on HN going back and forth about how outright lies by the system indight the whole thing for everybody, we should be careful to note that it's not for everybody, or rather, it's a known limitation so don't trust it to cite real cases for you as a lawyer, but using it to help you figure out what mens rea means in a practical sense by asking it questions about the concept, totally.

Honestly, hallucinations happen so rarely for me because of the kinds of things I ask it, that it doesn't happen enough for me to not believe it's answers in low-stakes situations, or situations on the level of horoscope generation, and I'm sure I'm not alone in treating ChatGPT that way, despite evidence to the contrary.

> …it is pretty sad since this is not a matter of taste or political inclination: there must be a single truth

This is more of a salient point that you perhaps realized. In life there is no single absolute, unknowable truth. Philosophy has spent the entire span of human existence grappling with this topic. The real risk with AI is not that we build some humanity-destroying AGI, but that we build a machine that is 'convincing enough' — and the idea that such a machine would be built by people who believe in objective truth is the most worrying part.

Depends, if you're a realist [1] (like most) then there can be such a thing as absolute truth, that you may not always be able to access.

[1] https://en.wikipedia.org/wiki/Philosophical_realism?wprov=sf...

This is teleologically false.

A teleological argument that assumes truth is contingent upon a specific worldview would indeed be flawed, because it would make truth an artifact of a given perspective rather than something independent of it.

> At the same time, part of the scientific community continues to diminish what was accomplished

Revisit the idea: part of the public is bewildered by voices that started calling "intelligence" what was and apparently still is the precise implementation of unintelligence. The fault is in some, many people - as usual.

Very recent state-of-the-art LLM models themselves declare that if the majority of their training data states that entity E is red they will say it's red, and if the majority says it's blue then they will say it's blue: that is the implementation of an artificial moron.

And in fact, very recent state-of-the-art LLM models state cretinous ideas that are child level - because "that's what they have heard" (stuck, moreover analytically, in the simplifications intrinsic in expression).

This architectural fault should be the foremost concern.

Psychological denial of what exactly? And what part of the article/preprints you are commenting on?

Every time an article exposing some limitation of the current wave of LLMs is submitted to HN there are comments like yours and I genuinely cannot understand the point you are trying to make: There is no such thing as a perfect technology, everything has limitations, and we can only improve our current state of the art by studying these and iterate.

I think if we referred to LLMs as AK (Artifical Knowledge) instead of AI it would be easier to have more cohesive discussions.

I don’t see how there can be a single truth when there is not even a single definition of many of the underlying terms (intelligence, AGI, etc) which this discipline supposedly defines itself by. Combine that with a lot of people with little philosophical perspective suddenly being confronted with philosophical topics and you end up with a discourse that personally I’ve mostly given up on participating in until things calm down again.

It feels like nobody remembers all the timelines for which we were supposed to have self-driving cars.

you are I think badly misrepresenting what Yann Le Cun said: he didn't say LLM's were a dead end, he said to do research in directions that do not require billions of dollars of investment to show results, in particular for PhD's this is sensible, and in view of recent cheaper results, prescient

Sensible with the caveat that deepseek R1 still took millions of dollars off compute time, so you're not training the next one on the box in your basement with a pair of 3090s (though you could certainly fine-tune a shared quantized model). you can't run the full sized model on anything cheap, so. basement researcher still need access to a decent amount of funding, which likely requires outside help.

It is becoming more and more important to determine for ourselves what is true and what is not. No person is right on most things, even when they are an expert in that thing. The biggest trap, is to believe someone because they are passionate, that they say it with conviction. Ignore most of the out of band signaling, take what they are saying and then also see if you can corroborate with another source.

There are so many people who are wrong about so may things.

I really appreciate that you are making your dev with ai videos, it shows people different, more humanistic ways of operating with AI.

Most of what I use AI for is to understand and relearn things I only thought I knew. This I think, is the most powerful use of AI, not in the code writing or the image generation, but in understanding and synthesis.

There is that hilarious tautological statement, "it is easy if you know it".

This video https://www.youtube.com/watch?v=TPLPpz6dD3A shows how to use AI to be a personal tutor using the Socratic Method. This is what people should be using AI for, have it test you for things you think you are already good at and you will find huge gaps in your own understanding. Now go apply it to things you have no clue about.

Speaking of parrots, a large volume of the anti AI sentiment, even here is by people repeating half truths they don't understand, confidently, about what AI cannot do. One would need a pretty tight formal case to prove such things.

Everyone should be playing, learning and exploring with these new tools, not shutting each other down.

Another trope that stands out is that someone will take a model, run a battery of tests against it and then make general statements about what LLMs can and cannot do without understanding their architecture, the training data, and the training itself.

And then they dress it up to sound scientific, when really they are making hasty generalizations to support a preconceived bias.

Yes, the stochastic parrots story is one of the most strong instances in recent times where experts in a field are made blind by their own expertise (the mental model they have of certain things) to the point of being incapable of seeing trivial evidences.

There’s a certain irony in hearing someone describe an LLM as a "stochastic parrot" for the ten-thousandth time when the only reason they’re doing so is that they’ve seen a sufficient number of other people using the exact same term (so now it's in their proverbial training data).

All they do is predict the next word!

It's an accurate description, though. People are running around convincing themselves that the RNG text generator is alive and it's worth slapping some sense into them.

it's no longer an accurate description.

with recent models, while stitching together training data is still most of the story, there's obviously something more going on, especially with regards to code, mathematics, and with the ability to successfully produce dialogue that respects pragmatics. this has nothing to do with a real thinking being, but it's become different from a stochastic parrot.

there are deluded people who think that Claude and ChatGPT are really thinking and feeling. it is admirable to try to argue them out of this position.

but if your argument involves saying things that are not true, and anyone can easily see are not true by interacting with a consumer-grade chatbot for 10 minutes, it will just discredit you, and those people will go on believing they're talking to HAL-9000.

I'm sure that there are people that are deluded into thinking ChatGPT loves them like a real life flesh and blood being can even when it says it can't, but we have such limited vocabulary, especially as laymen, for describing any non-human intelligence, that saying it's thinking and reasoning aren't entirely unreasonable words to describe what it's doing. sure, it's not thinking in the same way a human would, but when a computer, pre-LLM, and even pre-Internet, is doing something that requires the user to wait, saying the computer is "thinking" is an entirely accepted practice.

So if we want to get people to stop using the words thinking and reasoning, we have to get replacement words into the lexicon. if I tell an LLM A implies B implies C, and I tell it A is true and it's able to tell me that C is thus also true, sure it's entirely due to that much logic existing in its training corpus, but unless we get to a point where I can say that ChatGPT is dot-producting an essay for me, or some other phrase, saying it's not doing "thinking" is going to fall flat on its face. Hell, Deepseek R1's output for the local model literally says <think>. It may not be thinking in a biological being sense, and it may not be reason in a biological sense to conclude C if A implies B implies C, if A is true, but we lack the common colloquial language to describe it otherwise.

completely agree. people say a classic chess engine is "thinking" when it goes deeper into the search tree, but nobody is confused about that. This colloquial language doesn't bother me so much.

But there really are people who think they are talking to something more than that. Like within a conversation with today's consumer product, they sincerely believe that an actual being is instantiated who has goals and intentions and talks to them.

Using language like "thinking" doesn't bother me. I'm not a stickler for precision language in colloquial speech.

I do think it's important to deflate the hype and give context to what you mean by "thinking" in products, technologies and so on. Calling it a "stochastic parrot" is a bit pithy but not unreasonable. Plus it's memorable.

"Obviously" is doing some heavy lifting there. I don't think it's obvious at all.

What is obvious is people are seeing patterns in the output where there are none. We're hardwired for pattern-seeking and we're great at it. We see faces in clouds and hear words in static.

I've seen a number of people wax about how they interrogate ChatGPT until it gives them "the good stuff" and they're all just Clever Hans-ing themselves.

so here's an example: I saw the following amusing math joke on BlueSky, posted a couple days ago.

https://bsky.app/profile/elliotlip.bsky.social/post/3lh5bs7g...

"It's wild that I took a whole-ass class all about differentiable functions from R^2 to itself that satisfy one specific PDE."

output from Claude (one-shot, no retries or trickery) when asked to "explain the joke":

"The joke is playing off the fact that "complex analysis" (the study of differentiable functions satisfying the Cauchy-Riemann equations - a type of PDE) sounds like it should be more complex than it is when described technically. The humor comes from reducing an entire mathematical field to "functions that satisfy one specific PDE," making it sound surprisingly narrow or trivial despite its actual depth and importance."

Wikipedia and lots of math textbooks are in the training data, so it's not so impressive that this model will produce correct statements about math.

But the joke itself was not in the training data (to the best of my knowledge, it's this guy's original joke). And complex analysis was not mentioned in the joke. Yet somehow the text generated is correct with respect to both of those things.

I see things like this quite regularly, which under the "stochastic parrots" story, can't happen.

I've tried to phrase all these sentences very carefully to not claim there is any "agent" or "intelligence" behind the Claude product. There are many explanations for how a language model like this could imitate intelligent dialogue in ways that are somewhat fake and don't generalize -- I think this is what's happening. I also see things break down all the time and the sleight-of-hand fall apart. However, it is not "stochastic parrots" any more.

But what for? Human learning is becoming of diminishing utility as the machines improve. For example, I am now able to create computer programs and beautiful artwork without taking the time to master these skills. You could say that I can use art and programming as basic tools to accelerate my learning of bigger things, but whatever that bigger thing is AI is coming for it too. I can't imagine the progress the machines will achieve in 10 years. We'll be replaced.

The reason you overestimate their capabilities is because you use them for things you don’t know anything about. It’s like when your nephew made a simple HTML website for himself twenty years ago that was <h1>Hi I am Mark</h1> — it seemed impressive, but you just didn’t know that it wasn’t. Using LLMs in real world complex cases (in programming or art) instantly reveal their significant shortcomings. They are a very good nephew for making stuff that seem impressive, but a bad expert or consultant.

I'm sorry but they don't "do progress that was anticipated impossible", especially not every two months.

They were predicted to end the software engineering profession for almost four years already. And it just doesn't happen, even though they can bang out a perfect to-do list in React in a matter of seconds.

LLMs have incremental improvements on the quality of their responses as measured by benchmarks. The speed and cost of inference has also been improving. Despite that there was no major breakthrough since GPT 3.

People keep trying to make them reason, and keep failing at it.

> They were predicted to end the software engineering profession for almost four years already

ChatGPT was launched on November 30 2022. Two years and two months ago. The fact that in such a short timeframe you're talking about missed predictions is absurd, but telling of the accelerated timeframe in which we're living. The fact is that currently AI and LLMs are going through a phase of explosive improvement, to the point we can expect enormous improvements in capabilities every six months or so.

And what has enormously improved since ChatGPTs launch? Maybe you should ask it what it "thinks" about the hype surrounding it.

This is a good question. According to some the growth is exponential. Others think ChatGPT is basically still the same as it was at the end of 2022, minor differences aside. Why are the perspectives so different?

One of these archetypes is drowning in a hype-fueled news cycle, they mistake speculation for inevitability, dismiss skepticism as ignorance, and construct a reality where the technology's success is unquestionable.

The other is simply using the technology for what it's good for, observing that it's slowly, incrementally improving at tasks that it was already capable of since the major breakthrough, and acknowledging its limitations.

Incremental improvements don't give us any assurance that another major breakthrough is waiting around the corner.

ChatGPT launched with GPT-3.5. We're now at o1 and o3-mini and DeepSeek-R1, but even in the last year with GPT-4 and GPT-4o, it became better than almost everything involving text than average human. It writes better than average person, faster and cheaper. It parses unstructured data better than average person. There's large number of everyday tasks for which it's perfectly reasonable today to just throw them at ChatGPT. That's all last 1 to 1.5 years.

Oh, we have new letter-number combinations now. That is amazing. I stand corrected.

If you haven't tried using them then I am not sure your opinion on them is any good.

I dare say I'm more familiar with the capabilities of the leading models than certain big tech CEOs are, at least judging by their publicly communicated opinions.

I use the 4o very often in my work and it mostly sucks. Sometimes it’s very good, sometimes it has nice knowledge that was faster to find from it than a search engine. Mostly it spouts out unhelpful noise (for my problems).

I’m sure if you need to make a to-do list in react it’s like magic (until the app gets complicated). In real world use, not so much.

(Also I have often code reviewed PRs from people who are heavy users and surprise surprise - their output is trash and very prone to bugs or being out of spec.)

I also think 4o sucks, but have you tried DeepSeek R1 (free on their website)? I thought it night and day between 4o and o3-mini on the following topics:

- reverse engineering: when fed assembly (or decomp or mock impl), it's been consistently been able to figure out what the function actually does/why it's there from a high-level perspective. Whereas ChatGPT merely states the obvious

- very technical C++ questions: DSR1 gives much more detailed answers, with bullet points and examples. Much better writing style. Slightly prone to hallucinations, but not that much

- any controversial topic: ChatGPT models are trained to avoid these because of its "safety" training

ChatGPT is a bit better (and faster) at writing simple code and doing some math faster, but that's it.

(obviously, common sense about what to share and not to share with these chatbots still apply, etc.)

You can access DeepSeek R1? For me, both chat and API have been down for over a week now (it shut down minutes after I topped up my account and generated an API key - I never got to use it :/).

There's lots of fiddling with these models. I found Claude 3.5 Sonnet to be superior to both GPT-4o and o1-preview in around 99% of the things I do; I only started comparing it against o3-mini, and right now it's a mixed bag. Then again, I tend to develop and refine specific prompts for Sonnet, which I haven't for o1-preview and o3-mini, so that could be a factor. Etc.

> You can access DeepSeek R1?

Yes, well, I live in the EU and thus can avoid US work hours and Chinese peak hours. I think availability has been a bit better since they disabled websearch (also I noticed DSR1 half a week before it made the mainstream news).

> There's lots of fiddling with these models.

Agreed

I live in the EU too. For me, the status page[0] shows a continuous API outage for the past 8 days, that is still ongoing. Since it started, my API requests bounce back with an error, which changes seemingly at random between "unauthorized" and "insufficient balance". Neither of those reasons are valid, since I'm using a valid API key I made after creating an account, which I topped up with $20 (and have an invoice from them to prove it). I must have had a mightily bad luck that the service went down soon after I generated the API key - I'm guessing my user/key is currently stuck in the middle of some migration, or possibly wasn't captured in a backup and got subsequently wiped. For now, I'll just patiently wait for them to fix their service.

[0] - https://status.deepseek.com/

AFAIK it's hosted on Chutes for free too (though limited to between 2k and 10k output tokens). Azure as well, though it might be ratelimited there (or at least it is through openrouter)

If you don’t see the difference in quality of responses between GPT-3.5 as it launched in 2022 and o1/o3 then I don’t know what to tell you. I am using these models daily and the difference is night and day.

In addition to passing bar exam[0], improved performance on medical questions[1], economics questions that experts thought it was years away from[2], all the other things marked in green on page 6 were just the changes from 3.5 to 4: https://arxiv.org/pdf/2303.08774

4o added image analysis.

The o-series starting at o1 improves on 4o as per the margins in these charts: https://openai.com/index/learning-to-reason-with-llms/

I'll have to wait and see about o3, because only the mini model is out yet.

[0] https://law.stanford.edu/2023/04/19/gpt-4-passes-the-bar-exa...

[1] https://ai.nejm.org/doi/full/10.1056/AIdbp2300192

[2] https://www.betonit.ai/p/gpt-4-takes-a-new-midterm-and-gets

At this point just paste my comments into ChatGPT and ask it to explain to you what I mean by them. Then paste your response and ask it why it's not addressing the point made. At least use the tool for what it's good for.

So you're saying that it understands you better than I do?

I get that feeling too (in both directions) but this vague and hard to quantify sensation is not what I'd suggest in response to your clearly stated question:

> And what has enormously improved since ChatGPTs launch?

Which is, I think, answered by the things I listed.

It doesn’t understand me, but it could help you understand. What you listed aren’t major unexpected leaps but incremental improvements on things that already were known to be possible.

But you insist on being obstinate. ChatGPT advised me to disengage from this conversation.

This is highly misleading.

ChatGPT did not ace the bar exam -- it was basically percentile graded against a group of people who mostly failed. If compared to real lawyers, it was 15th percentile on the essay portion

[0] https://law-ai.org/re-evaluating-gpt-4s-bar-exam-performance...

I said pass, not ace.

15th percentile of passes, on the weakest aspect, is still a big improvement over "not passing". That improvement is what I wish to highlight.

(The observation that 48th percentile (lowest overall from your link, let alone 15th for essays) of passes corresponds to 90th percentile of all exam takers, suggests that perhaps too many humans are taking the exams before they're ready).

> to the point we can expect enormous improvements in capabilities every six months or so

Not really, we just can see we've had improvements. That is not evidence of upcoming improvement.

I use LLM’s daily so I’m no skeptic. We are not seeing enormous improvements every 6 months, that’s hyperbolic. There has been a significant improvement since GPT 3.5, I’ll give you that, but even in those ~2 years I don’t think I’d describe the improvement as “enormous”. The capabilities are similar with output quality improving by a noticeable degree.

OpenAI API for GPT-3 was launched on June 11, 2020, that's four years and seven months ago:

https://news.ycombinator.com/item?id=23489653

I used that API. It was literally autocomplete — if you wanted it to answer a question, you had to start with, say:

Translate into German:

Q: "Good Morning" A: "Guten Morgen"

Q: "<the thing you actually wanted to translate>

And even then it might answer with "Good Morning" in five different languages.

InstructGPT is what turned GPT-3 into the ChatGPT-3.5 model.

Note that GP said "since GPT 3" but the parent responded with "ChatGPT". My response was to clarify the timeframe that has elapsed since GPT-3.

GPT-2 was released November, 2019, so that's five years ago. GPT-3 isn't the advancement to look at, but ChatGPT, which didn't use GPT-3, but a newer RLHF'd model, based off GPT-3, has the 2022 launch date.

https://news.ycombinator.com/item?id=21454273

LLMs can hammer out existing solutions to problems, but not never before seen problems.

SE is a good example - I get a lot of help from LLM tools and I think we're learning how to use them better across realistic SDLC processes as well, but we're not replacing lots of people at the moment. On the other hand I saw a business case from one of the big SI's (not my employer but in a deck that was shown by the SI in an discussion) that described the need to move their Indian software dev workforce from 350k FTE to 50K FTE over the next five years.

I think that the onshore impacts will be much lower or negligible, or possibly even positive, because so much work has been offshored already, and as is well worn in every discussion, Jevons paradox may drive up demand significantly (to be fair I believe this as wherever I have worked we've had 3x+ demand (with business cases) for development projects and had to arbitrarily cull 2x of it at the beginning of each year. So, just like the 30 people in India that are working on my project won't do anything useful unless we feed the work to them, the LLM's won't do anything useful either. And just like we have to send lots of work back to India because it's not right, the same is true of LLM's. The difference is that I won't spend 4 hrs on a friday afternoon on Teams discussing it.

But this is not surprising because we've had big impacts from tools like IDE's, VM's, and compilers which have driven seismic changes in our profession, I think that LLM's are just another one of those.

What I'm watching for is an impact in a non tech domain like healthcare or social care. These are important domains that are overwhelmed with demand and riddled with makework, yet so far LLM's have made very little impact. At least, I am not seeing health insurance rates being cut, hospital waiting lists fall or money and staff being redeployed from back office functions to front line functions.

Why hasn't this started?

Progress is what happens thanks to AI skeptics busy defining model limitations. The limitations set attractive bars to pass.

Did you read the article? Dziri and Peng are not the “skeptical AI community,” they are in fact die hard AI researchers. This is like saying people who run benchmarks to find performance problems in code are skeptics or haters.

I read the article: it does not look like very good research: It's simple to find flaws in LLMs reasoning / compositional capabilities looking at problems that are at the limit of what they can do now, or just picking problems that are very far from their computational model, or submitting riddles. But there is no good analysis of the limitations, nor inspection of how/how much better recently LLMs got exactly at this kind of problems. Also the article is full of uninformative and obvious things to show how LLMs fail in stupid tasks such as multiplication between large numbers.

But the most absurd thing is that the paper looks at computational complexity in terms of direct function composition, and there is no reason an LLM should just use this kind of model when emitting many tokens. Note that even when CoT is not explicit, the LLM output that starts to shape the thinking process still makes it able to have technically unbound layers. With CoT this is even more obvious.

Basically there is no bridge between their restricted model and an LLM.

I think that “part of the scientific community” actually wants to do what needs to be done: “We have to really understand what’s going on under the hood,” she said. “If we crack how they perform a task and how they reason, we can probably fix them. But if we don’t know, that’s where it’s really hard to do anything.”

Well, there appears to be evolution in human perception of capabilities of LLMs. An example, the 'stochastic parrots' notion seems to have mostly died out, at least in HN comments.

Reply on news.ycombinator.com