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

Mira Murati leaves OpenAI

twitter.com

Post reply on HN
Print view View original

531–540 of 644 posts

Re: Mira Murati leaves OpenAI

#531

Post by bmitc »

Sunhold wrote: post #529
bmitc wrote: post #524

Earlier quoted context omitted.

I continue to be surprised by the talk of general artifical intelligence when it comes to LLMs. At their core, they are text predictors, and they're often pretty good at that. But anything beyond that, they are decidely unimpressive. I use Copilot on a daily basis, which uses GPT 4 in the backend. It's wrong so often that I only really use it for boilerplate autocomplete, which I still have to review. I've had collea…

Look at the sample chain-of-thought for o1-preview under this blog post, for decoding "oyekaijzdf aaptcg suaokybhai ouow aqht mynznvaatzacdfoulxxz". At this point, I think the "fancy autocomplete" comparisons are getting a little untenable. https://openai.com/index/learning-to-reason-with-llms/

How exactly does a blog post from OpenAI about a preview release address my comment or make fancy autocomplete comparisons untenable?

Re: Mira Murati leaves OpenAI

#532

Post by HarHarVeryFunny »

svara wrote: post #445
HarHarVeryFunny wrote: post #383

Earlier quoted context omitted.

> There's a huge amount of circuitry between the input and the output of the model Yeah - but it's just a stack of transformer layers. No looping, no memory, no self-modification (learning). Also, no magic.

No looping, but you can unroll loops to a fixed depth and apply the model iteratively. There obviously is memory and learning. Neuroscience hasn't found the magic dust in our brains yet, either. ;)

Zero memory inside the model from one input (ie token output) to the next (only the KV cache, which is just an optimization). The only "memory" is what the model outputs and therefore gets to re-consume (and even there it's an odd sort of memory since the model itself didn't exactly choose what to output - that's a random top-N sampling).

There is no real runtime learning - certainly no weight updates. The weights are all derived from pre-training, and so the runtime model just represents a frozen chunk of learning. Maybe you are thinking of "in-context learning", which doesn't update the weights, but is rather the ability of the model to use whatever is in the context, including having that "reinforced" by repetition. This is all a poor substitute for what an animal does - continuously learning from experience and exploration.

The "magic dust" in our brains, relative to LLMs, is just a more advanced and structure architecture, and operational dynamics. e.g. We've got the thalamo-cortical loop, massive amounts of top-down feedback for incremental learning from prediction failure, working memory, innate drives such as curiosity (prediction uncertainty) and boredom to drive exploration and learning, etc, etc. No magic, just architecture.

Re: Mira Murati leaves OpenAI

#533

Post by freilanzer »

yas_hmaheshwari wrote: post #471
bansheeps wrote: post #326

Update: Looks like Barret Zoph, GPT-4's post training (co-)lead is also leaving: https://x.com/barret_zoph/status/1839095143397515452

Whoa! This definitely looks much more troubling for the company now. Can't decide it is because AGI is coming very soon OR AGI is very far away

Why would they leave if AGI was near?

Re: Mira Murati leaves OpenAI

#534

Post by Sunhold »

bmitc wrote: post #531
Sunhold wrote: post #529

Earlier quoted context omitted.

Look at the sample chain-of-thought for o1-preview under this blog post, for decoding "oyekaijzdf aaptcg suaokybhai ouow aqht mynznvaatzacdfoulxxz". At this point, I think the "fancy autocomplete" comparisons are getting a little untenable. https://openai.com/index/learning-to-reason-with-llms/

How exactly does a blog post from OpenAI about a preview release address my comment or make fancy autocomplete comparisons untenable?

It shows that the LLM is capable of reasoning.

Re: Mira Murati leaves OpenAI

#535

Post by riazrizvi »

yas_hmaheshwari wrote: post #471
bansheeps wrote: post #326

Update: Looks like Barret Zoph, GPT-4's post training (co-)lead is also leaving: https://x.com/barret_zoph/status/1839095143397515452

Whoa! This definitely looks much more troubling for the company now. Can't decide it is because AGI is coming very soon OR AGI is very far away

Seems obvious to me that the quality of the models is not improving since GPT-4. The departures I’m guessing are a problem talent has with ‘founder mode’, Altman’s choice of fast pace, this absence of model improvement with these new releases, and the relative temptation of personal profit outside of OpenAI’s not-for-profit business model. People think they can do better in control themselves. I suspect they are all under siege with juicy offers of funding and opportunities. Whether or not they will do better is another story. My money is on Altman, I think he is right on the dumpster rocket idea, but it’s very difficult to see that when you’re a rocket scientist.

Re: Mira Murati leaves OpenAI

#536

Post by HeatrayEnjoyer »

HarHarVeryFunny wrote: post #383
svara wrote: post #222

Earlier quoted context omitted.

I truly think you haven't really thought this through. There's a huge amount of circuitry between the input and the output of the model. How do you know what it does or doesn't do? Humans brains "just" output the next couple milliseconds of muscle activation, given sensory input and internal state. Edit: Interestingly, this is getting downvotes even though 1) my last sentence is a precise and accurate statement of th…

> There's a huge amount of circuitry between the input and the output of the model Yeah - but it's just a stack of transformer layers. No looping, no memory, no self-modification (learning). Also, no magic.

>no memory, no self-modification (learning).

This is also true of those with advanced Alzheimer's disease. Are they not conscious as well? If we believe they are conscious then memory and learning must not be essential ingredients.

Re: Mira Murati leaves OpenAI

#537

Post by ToucanLoucan »

Sunhold wrote: post #529
bmitc wrote: post #524

Earlier quoted context omitted.

I continue to be surprised by the talk of general artifical intelligence when it comes to LLMs. At their core, they are text predictors, and they're often pretty good at that. But anything beyond that, they are decidely unimpressive. I use Copilot on a daily basis, which uses GPT 4 in the backend. It's wrong so often that I only really use it for boilerplate autocomplete, which I still have to review. I've had collea…

Look at the sample chain-of-thought for o1-preview under this blog post, for decoding "oyekaijzdf aaptcg suaokybhai ouow aqht mynznvaatzacdfoulxxz". At this point, I think the "fancy autocomplete" comparisons are getting a little untenable. https://openai.com/index/learning-to-reason-with-llms/

I’m not seeing anything convincing here. OpenAI says that it’s models are better at reasoning and asserts they are testing this by comparing how it does solving some problems between o1 and “experts” but it doesn’t show the experts or o1s responses to these questions nor does it even deign to share what the problems are. And, crucially, it doesn’t specify if writings on these subjects were part of training data.

Call me a cynic here but I just don’t find it too compelling to read about OpenAI being excited about how smart OpenAIs smart AI is in a test designed by OpenAI and run by OpenAI.

Re: Mira Murati leaves OpenAI

#538

Post by HarHarVeryFunny »

Sunhold wrote: post #529
bmitc wrote: post #524

Earlier quoted context omitted.

I continue to be surprised by the talk of general artifical intelligence when it comes to LLMs. At their core, they are text predictors, and they're often pretty good at that. But anything beyond that, they are decidely unimpressive. I use Copilot on a daily basis, which uses GPT 4 in the backend. It's wrong so often that I only really use it for boilerplate autocomplete, which I still have to review. I've had collea…

Look at the sample chain-of-thought for o1-preview under this blog post, for decoding "oyekaijzdf aaptcg suaokybhai ouow aqht mynznvaatzacdfoulxxz". At this point, I think the "fancy autocomplete" comparisons are getting a little untenable. https://openai.com/index/learning-to-reason-with-llms/

It depends on how well you understand how the fancy autocomplete is working under the hood.

You could compare GPT-o1 chain of thought to something like IBM's DeepBlue chess-playing computer, which used MTCS (tree search, same as more modern game engines such as AlphaGo)... at the end of the day it's just using built-in knowledge (pre-training) to predict what move would most likely be made by a winning player. It's not unreasonable to characterize this as "fancy autocomplete".

In the case of an LLM, given that the model was trained with the singular goal of autocomplete (i.e. mimicking the training data), it seems highly appropriate to call that autocomplete, even though that obviously includes mimicking training data that came from a far more general intelligence than the LLM itself.

All GPT-o1 is adding beyond the base LLM fancy autocomplete is an MTCS-like exploration of possible continuations. GPT-o1's ability to solve complex math problems is not much different from DeepBlue's ability to beat Garry Kasparov. Call it intelligent if you want, but better to do so with an understanding of what's really under the hood, and therefore what it can't do as well as what it can.

Re: Mira Murati leaves OpenAI

#539

Post by bamboozled »

freilanzer wrote: post #533
yas_hmaheshwari wrote: post #471

Earlier quoted context omitted.

Whoa! This definitely looks much more troubling for the company now. Can't decide it is because AGI is coming very soon OR AGI is very far away

Why would they leave if AGI was near?

Enjoy their life while they can?
Post reply on HN