Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

OpenAI, Google and Anthropic are struggling to build more advanced AI

bloomberg.com

Post reply on HN
Print view View original

341–350 of 622 posts

Re: OpenAI, Google and Anthropic are struggling to build more advanced AI

#341

Post by saalweachter »

ppeetteerr wrote: post #250
crystal_revenge wrote: post #127

Earlier quoted context omitted.

I don't think we've even started to get the most value out of current gen LLMs. For starters very few people are even looking at sampling which is a major part of the model performance. The theory behind these models so aggressively lags the engineering that I suspect there are many major improvements to be found just by understanding a bit more about what these models are really doing and making re-designs based on…

The reason people are holding out is that the current generation of models are still pretty poor in many areas. You can have it craft an email, or to review your email, but I wouldn't trust an LLM with anything mission-critical. The accuracy of the generated output is too low be trusted in most practical applications.

Any email you trust an LLM to write is one you probably don't need to send.

Re: OpenAI, Google and Anthropic are struggling to build more advanced AI

#343

Post by sssilver »

One thing that makes the established AIs less ideal for my (programming) use-case is that the technologies I use quickly evolve past whatever the published models "learn".

On the other hand, a lot of these frameworks and languages have relatively decent and detailed documentation.

Perhaps this is a naive question, but why can't I as a user just purchase "AI software" that comes with a large pre-trained model to which I can say, on my own machine, "go read this documentation and help me write this app in this next version of Leptos", and it would augment its existing model with this new "knowledge".

Re: OpenAI, Google and Anthropic are struggling to build more advanced AI

#345

Post by llm_trw »

delusional wrote: post #308
dmd wrote: post #166

Earlier quoted context omitted.

Pretty much, yes. My job is pretty fun; it mostly entails things like "take this horrible file workflow some research assistant came up with while high 15 years ago and turn it into a newer horrible file format a NEW research assistant came up with (also while high) 3 years ago" - and automate this in our data processing pipeline.

If I understand that correctly you're converting file formats? That's not exactly "novel"

This is exactly the type of novel work that llms are good at. It's tedious and has annoying internal logic, but that logic is quite flat and there are a million examples to generalise from.

What they fail at is code with high cyclomatic complexity. Back in the llama 2 finetune days I wrote a script that would break down what each node in the control flow graph into its own prompt using literate programming and the results were amazing for the time. Using the same prompts I'd get correct code in every language I tried.

Re: OpenAI, Google and Anthropic are struggling to build more advanced AI

#346

Post by falcor84 »

xanderlewis wrote: post #323
david2ndaccount wrote: post #99

Earlier quoted context omitted.

Without a non-linear activation function, chaining perceptrons together is equivalent to one large perceptron.

Yep. falcor84: you’re thinking of the so-called ‘multilayer perceptron’ which is basically an archaic name for a (densely connected?) neural network. I was referring to traditional perceptrons.

While ReLU is relatively new, AI researchers have been aware of the need for nonlinear activation functions and building multilayer perceptrons with them since the late 1960s, so I had assumed that's what you meant.

Re: OpenAI, Google and Anthropic are struggling to build more advanced AI

#347

Post by famouswaffles »

HarHarVeryFunny wrote: post #284
famouswaffles wrote: post #216

Earlier quoted context omitted.

>Regardless of what Altman says, its been two years since OpenAI released GPT-4, and still no GPT-5 in sight. It's been 20 months since 4 was released. 3 was released 32 months after 2. The lack of a release by now in itself does not mean much of anything.

By itself, sure, but there are many sources all pointing to the same thing. Sutskever, recently ex. OpenAI, one of the first to believe in scaling, now says it is plateauing. Do OpenAI have something secret he was unaware of? I doubt it. FWIW, GPT-2 and GPT-3 were about a year apart (2019 "Language models are Unsupervised Multitask Learners" to 2020 "Language Models are Few-Shot Learners"). Dario Amodei recently said…

>Sutskever, recently ex. OpenAI, one of the first to believe in scaling, now says it is plateauing.

Blind scaling sure (for whatever reason)* but this is the same Sutskever who believes in ASI within a decade off the back of what we have today.

* Not like anyone is telling us any details. After all, Open AI and Microsoft are still trying to create a 100B data center.

In my opinion, there's a difference between scaling not working and scaling becoming increasingly infeasible. GPT-4 is something like x100 the compute of 3 (Same with 2>3).

All the drips we've had of 5 point to ~x10 of 4. Not small but very modest in comparison.

>FWIW, GPT-2 and GPT-3 were about a year apart (2019 "Language models are Unsupervised Multitask Learners" to 2020 "Language Models are Few-Shot Learners").

Ah sorry I meant 3 and 4.

>Dario Amodei recently said that with current gen models pre-training itself only takes a few months (then followed by post-training, etc). These are not year+ training runs.

You don't have to be training models the entire time. GPT-4 was done training in August 2022 according to Open AI and wouldn't be released for another 8 months. Why? Who knows.

Re: OpenAI, Google and Anthropic are struggling to build more advanced AI

#348

Post by moffkalast »

chongli wrote: post #334
hackinthebochs wrote: post #260

Earlier quoted context omitted.

>As soon as you ask it novel questions it falls apart. What do you mean by novel? Almost all sentences it is prompted on are brand new and it mostly responds sensibly. Surely there's some generalization going on.

Novel as in requiring novel reasoning to sort out. One of the classic ways to expose the issue is to take a common puzzle and introduce irrelevant details and perhaps trivialize the solution. LLMs pattern match on the general form of the puzzle and then wander down the garden path to an incorrect solution that no human would fall for. The sort of generalization these things can do seems to mostly be the trivial sort:…

Well the problem with that approach is that LLMs are still both incredibly dumb and small, at least compared to the what, 700T params of a human brain? Can't compare the two directly, especially when one has a massive recall advantage that skews the perception of that. But there is still some inteligence under there that's not just memorization. Not much, but some.

So if you present a novel problem it would need to be extremely simple, not something that you couldn't solve when drunk and half awake. Completely novel, but extremely simple. I think that's testable.

Re: OpenAI, Google and Anthropic are struggling to build more advanced AI

#349

Post by JohnMakin »

kenjackson wrote: post #299
JohnMakin wrote: post #84

Earlier quoted context omitted.

> is it any less believable than getting general intelligence by training a blob of meat? Yes, because we understand the rough biological processes that cause this, and they are not remotely similar to this technology. We can also observe it. There is no evidence that current approaches can make LLM's achieve AGI, nor do we even know what processes would cause that.

> because we understand the rough biological processes that cause this We don't have a rough understanding of the biological processes that cause this, unless you literally mean just the biological process and not how it actual impacts learning/intelligence. There's no evidence that we (brains) have achieved AGI, unless you tautologically define AGI as our brains.

> We don't have a rough understanding of the biological processes that cause this,

Yes we do. We know how neurons communicate, we know how they are formed, we have great evidence and clues as to how this evolved and how our various neurological symptoms are able to interact with the world. Is it a fully solved problem? no.

> unless you literally mean just the biological process and not how it actual impacts learning/intelligence.

Of course we have some understanding of this as well. There's tremendous bodies of study around this. We know which regions of the brain correlate to reasoning, fear, planning, etc. We know when these regions are damaged or removed what happens, enough to point to a region of the brain and say "HERE." That's far, far beyond what we know about the innards of LLM's.

> here's no evidence that we (brains) have achieved AGI, unless you tautologically define AGI as our brains.

This is extremely circular because the current definition(s) of AGI always define it in terms of human intelligence. Unless you're saying that intelligence comes from somewhere other than our brains.

Anyway, the brain is not like a LLM, in function or form, so this debate is extremely silly to me.

Re: OpenAI, Google and Anthropic are struggling to build more advanced AI

#350

Post by hackinthebochs »

chongli wrote: post #334
hackinthebochs wrote: post #260

Earlier quoted context omitted.

>As soon as you ask it novel questions it falls apart. What do you mean by novel? Almost all sentences it is prompted on are brand new and it mostly responds sensibly. Surely there's some generalization going on.

Novel as in requiring novel reasoning to sort out. One of the classic ways to expose the issue is to take a common puzzle and introduce irrelevant details and perhaps trivialize the solution. LLMs pattern match on the general form of the puzzle and then wander down the garden path to an incorrect solution that no human would fall for. The sort of generalization these things can do seems to mostly be the trivial sort:…

Why is your criteria for "on the path towards AGI" so absolutist? For it to be on the path towards AGI and not simply AGI it has to be deficient in some way. Why does the current failure modes tell you its on the wrong path? Yes, it has some interesting failure modes. The failure mode you mention is in fact very similar to human failure modes. We very much are prone to substituting the expected pattern when presented with a 99% match to a pattern previously seen. They also have a lot of inhuman failure modes as well. But so what, they aren't human. Their training regimes are very dissimilar to ours and so we should expect some alien failure modes owing to this. This doesn't strike me as good reason to think they're not on the path towards AGI.

Yes, LLMs aren't very good at reasoning and have weird failure modes. But why is this evidence that its on the wrong path, and not that it just needs more development that builds on prior successes?

Post reply on HN