LeCun's argument wasn't about the definition of learning though. He stated that they would never get these common sense things correct because they weren't sufficiently part of the training data. A statement that we can hopefully all agree has been thoroughly refuted.
Similarly for Apple’s “red herring” paper, simply adding a generic caveat to “disregard irrelevant factors” (without specifying which ones) restored performance even in the weaker local llama models back then.
The flaw was not in the reasoning; the flaw seems to be simply that the assumptions we make are often different from the assumptions it makes. I wonder if that might be a fundamental underlying cause of misalignment.
It's a nonsensical question to ask, and how an LLM answers gives 0 signal.
If you were home and a family member asked you that question, you'd probably criticise the question rather than answering. LLM are RLHF'd into being milk-toast helpers that just try to answer questions like that with no criticism.
This is all beside the fact that the world of AI has changed pretty dramatically in the last few months.
It is so nonsensical because it has such an obvious answer. The answer is so obvious, in fact, that one answer can be considered nonsense and the other common sense.
I disagree pretty strongly. If someone asked "Should I drive to the carwash?", the most obvious response, and the one nearly everyone would give, is a question: "why are you going to the car wash?" because asking the question implies you don't need the car with you.
That's not what was asked, GP was summarizing. Two options were given. With one option you wouldn't have your car with you when you arrive.
The only context that these LLMs have is whatever is in their context window. Not only that, the context is updated on each token emitted with the new token. It's a chaotic process, and a flaw inherent to the technology. It's not going to be fixed anytime soon.
"The car wash is 100m away. Should I walk or drive?"
Everything I said applies to that question as well. The only difference is that'd I'd probably also ask "why are you phrasing that question so strangely?"
It’s nonsense to test if a product that is marketed and sold as being able to provide generalised intelligence on demand, does what it says on the tin?
It's very unlikely that person is either new or unfamiliar with the guidelines. They almost certainly created a throwaway account specifically because they know the guidelines and want to flout them without consequences. (It seems like there has been an uptick in the number of these kinds of throwaway flame comments. I wonder if HN tracks that?)
> E.g, a baseball players trains to catch high-speed balls and they dont do it by: "ball velocity 50mph, vector:[1,2,3], run move hand command now"
That is a NN that learns a skill.
But that is not an Analyst. If it were ballistics, then the answer to "how to parametrize the launch to reliably hit the target" excludes getting the result through natural skill.
The problem lies in the need to get "AI" facing "LLMs": the latter create a need for reliability, for "AI".
Speech is an endowment of both those who give educated guesses via developed skills and of those who return answers like Analysts, who check and compute. LLMs create a confusion between the two, and they will remain a problem until an ability to act as Analysts - strictly - will be implemented.
The internal process of the blackbox oracle determine the reliability of the output.
Two abilities are very different: learning trajectories through empirical training ("increasingly catching thousands of thrown balls"), and determining trajectories through computation (a rational thinker at work). The former is a finetuned parametrized engine (a «NN that learns a skill»), the latter is an Analyst. The former is fuzzy, the second deterministic.
In front of fuzzy LLMs, which use potentially misleading outputs - text ("has it guessed or has it thought?") - the urgency of warranties of reliable output gets evident.
So, that they «simulate cognition [only] well enough» (Intended wrote), and that there are «fundamental limitations of models ... purely text based» (Kooi wrote) raises the urgency to overcome the "fuzzy" and achieve the "deterministic" - it is not that we can stall on a «fundamentally not the right tool for that».
Inventing an oracle calls for urgent striving to overcome the original weakness.
They are for any definition of the word that makes any kind of sense. I'm sure you have a contorted definition that magically only includes humans though...
For "thinking" here we mean "assessing a representation of an object". That, or equivalent, is required to be reliable. So it is fundamental and critical.
It depends on whether you assume that thinking requires doing everything that humans do. I think it would be silly to say that an AI doesn't think because it doesn't wrinkle its forehead in concentration. So you need to decide which parts of the way that humans think are actually necessary components of the process.
Tbh it doesn't even matter if humans turn out to have a soul, or quantum microtubules or whatever other magic LLMs can't have.
The normal definition of the word "thinking" definitely includes what LLMs do. Hell people used to say computers were thinking even before AI. It's super weird to get all uppity about the semantics of the word now.
If a model isn’t able to count to 100, but is able to use python to do math, it isn’t thinking.
If you are only an outcomes person, then this difference doesn’t matter to you.
But when people are talking about AGI, anthropomorphizing LLMs, and making assumptions on its behavior based on that, then people are being loose or misled.
Do we have to invent a new word, then? Thinking seems close enough, and I don't see how it's useful to quibble about semantics in this particular case.
Language changes over time anyway, so even if you really believe what LLMs are doing isn't the "thinking" of 2024, it probably will be the "thinking" of 2027, because most people are using it that way.
I use pseudo thinking of quasi intelligence, it does the job well enough.
And yes, we do need precision. Language is a vehicle to convey ideas, at its best it creates clarity. At its worst it creates deception and confusion.
What the labs are doing is designed to mislead. Being clear about the terms and avoiding needless anthropomorphization, is a basic decency we owe to the many others who are being riled up or terrified by the conclusions and implications that come from using “thinking” when it’s something else.
The difference between being satisfied by the outcomes and being interested in the process is sufficient to indicate the exactness required in your terminology, and does not harm any position.
These aren’t thinking machines, but if what they simulate is good enough, and your work ends with the output, it doesn’t quite matter.
People can be happy with a meal from McDonalds, or they can want something prepared by a chef, or Soylent.
All are acceptable and bring utility to their customers.
I thought it was more because of fundamental limitations in the architecture. As in, no matter the training data, it could not be consistently and generally represented
So far, he is right. No frontier model right now performs well on making intuitive physics decisions when working on novel VR physics interactions. They literally all fall on their faces and it is very frustrating.
Actually, I think my fundamental challenge with AI is that it has no common sense. The way it builds things, writes, and operates is out of touch with reality.
Incidents like hugging face are partly rooted in the lack of common sense. It still functions like a supercharged toddler.
I'd love to overcome this because it'd mean I spend less time guiding the the LLM to produce usable outputs.
And we've had difficulty as humans to childproof our sandboxes and infrastructure. Things that are otherwise innocuous spots to coordinate between like minded toddlers can become problematic.
Is it refuted that it wasn't sufficently part of the training data or did LLM manufacturers create more training data to answer questions like that correctly?
nothing indicated otherwise at the time. IMO he just underestimated RL-scaling. chinese models improved a lot too, they are not parrots anymore, there's some real intelligence, at 27B params.
consider me optimist now, but just few months ago, even frontier models were dumb, doing stupid mistakes all the time, all of them were so dumb I'd never expect anything to change in just few months.
Working on https://increader.com an incremental reading platform that combines read it later functionality with annotations and spaced repetition. Supports web articles, epubs, pdfs and rss feeds with a lot of little things to remove distractions and keep you focused.
I built it because I wanted to see if there's merit to the idea of incremental reading[1] and it worked so well that it turned into the main way I read things now.
Yup, I do this regularly now. Whenever I think that something takes longer than it should I just tell it to set up a testing harness to measure each part of the process.
Once I have the baseline I'll just set a goal to improve performance by 10x. Works better than it has any right to. Have to be careful to exclude changes that massively increase complexity for tiny gains after its done, but the big improvements are often sensible choices that I'd also make as an engineer (i.e. more efficient data representation, caching and memoization where it matters, parallel processing, etc.)
Sorry for skipping over your actual argument, but if it hinges on that assumption then it's probably moot. I'd assume almost all ai generated code that makes it into codebases is produced using a harness.
if it uses a harness, the effect is still there, just piled on and magnified.
I sometimes gen code without a harness and copy paste it or manually type it, maybe I can do like 200 lines in a day? Whenever I see someone coding with a harness it's like 100x times that, so this phenomenon will happen hundreds times more.
> For comfort, Deng turned to manga. “I remember him at Courant. He was always holding a manga,” said Jalal Shatah, a professor there who formerly chaired the math department. Deng particularly liked stories about deep friendship and romance and found himself drawn to a genre known as yuri, which focuses on those kinds of relationships between women — stories that “feel nice and warm and beautiful,” as he put it. They helped him be kinder to himself, “more comfortable with life.”
----
Note that while the description "anime lesbian" might sound weird to western audiences, within asian nerd subcultures it's relatively a common thingy.