News for hackers. Stuff that matters.
Re:OpenAI co-founder John Schulman says he will leave and jo... by Geezus_42 on Tuesday August 06, 2024 @07:23AM (#41169743)
I still haven't seen it do anything actually interesting. Especially when you consider that you have fact check the AI.
Re:OpenAI co-founder John Schulman says he will leave and jo... by lambdaba on Tuesday August 06, 2024 @08:10AM (#41170086)
I'm continously baffled by such comments. Have you really tried? Especially newer models like Claude 3.5?
Re:OpenAI co-founder John Schulman says he will leave and jo... by dartos on Tuesday August 06, 2024 @08:50AM (#41170371)
I have, yeah.

Still useless for my day to day coding work.

Most useful for whipping up a quick bash or Python script that does some simple looping and file io.

Re:OpenAI co-founder John Schulman says he will leave and jo... by somenameforme on Tuesday August 06, 2024 @08:58AM (#41170441)
To be fair, LLMs are pretty good natural language search engines. Like when I'm looking for something in an API that does something I can describe in natural language, but not succinctly enough to show up in a web search, LLMs are extremely handy, at least when they don't just randomly hallucinate the API. On the other hand I think this is more of a condemnation of the fact that search tech has not 'really' meaningfully advanced beyond where it was 20 years ago, more than it is a praise of LLMs.
Re:OpenAI co-founder John Schulman says he will leave and jo... by dartos on Tuesday August 06, 2024 @09:05AM (#41170504)
> LLMs are extremely handy, at least when they don't just randomly hallucinate

I work in tech and it’s my hobby, so that’s what a lot of my googling goes towards.

LLMs hallucinate almost every time I ask them anything too specific, which at this point in my career is all I’m really looking for. The time it takes for me to realize an llm is wrong is usually not too bad, but it’s still time I could’ve saved by googling (or whatever trad search) for the docs or manual.

I really wish they were useful, but at least for my tasks they’re just a waste of time.

I really like them for quickly generating descriptions for my dnd settings, but even then they sound samey if I use them too much. Obviously they’d sound samey if I made up 20 at once too, but at that point I’m not really being helped or enhanced by using an LLM, it’s just faster at writing than I am.

Re:OpenAI co-founder John Schulman says he will leave and jo... by Workaccount2 on Tuesday August 06, 2024 @10:06AM (#41170996)
I don't mean this as a slight, just an observation I have seen many times - people who struggle with utility from SOTA LLM's tend to not have spent enough time with them to feel out good prompting. In the same way that there is a skill for googling information, there is a skill for teasing consistent good responses from LLM's.
Re:OpenAI co-founder John Schulman says he will leave and jo... by dartos on Tuesday August 06, 2024 @11:05AM (#41171578)
Why spend my time teasing and coaxing information out of a system which absolutely does make up nonsense when I can just read the manual?

I spent 2023 developing LLM powered chatbots with people who, purportedly, were very good at prompting, but never saw any better output than what I got for the tasks I’m interested in.

I think the “you need to get good at prompting” idea is very shallow. There’s really not much to learn about prompting. It’s all hacks and anecdotes which could change drastically from model to model.

None of which, from what I’ve seen, makes up for the limitations of LLM no matter how many times I try adding “your job depends on Formatting this correctly “ or reordering my prompt so that more relevant information is later, etc

Prompt engineering has improved RAG pipelines I’ve worked on though, just not anything in the realm of comprehension or planning of any amount of real complexity.

Re:OpenAI co-founder John Schulman says he will leave and jo... by danielbln on Tuesday August 06, 2024 @10:21AM (#41171112)
People also continue to use them as knowledge databases, despite that not being where they shine. Give enough context into the model (descriptions, code, documentation, ideas, examples) and have a dialog, that's where these strong LLMs really shine.
Re:OpenAI co-founder John Schulman says he will leave and jo... by dartos on Tuesday August 06, 2024 @11:12AM (#41171656)
Summarizing, doc qa, and unstructured text ingestion are the killer features I’ve seen.

The 3rd one still being quite involved, but leaps and bounds easier than 5 years ago.

Re:OpenAI co-founder John Schulman says he will leave and jo... by bamboozled on Tuesday August 06, 2024 @09:23AM (#41170652)
I hear a lot of people say good things about CoPilot too but I absolutely hate it. I have it enabled for some reason still, but it constantly suggests incorrect things. There has been a few amazing moments but man there is a lot of "bullshit" moments.
Re:OpenAI co-founder John Schulman says he will leave and jo... by Workaccount2 on Tuesday August 06, 2024 @10:00AM (#41170938)
Even when we get a gen AI that exceeds all human metrics, there will 100% still be people who with a straight face will say "Meh, I tried it and found it be pretty useless for my work."
Re:OpenAI co-founder John Schulman says he will leave and jo... by RHSman2 on Tuesday August 06, 2024 @07:43AM (#41169868)
It spends money really well.
Re:OpenAI co-founder John Schulman says he will leave and jo... by snapcaster on Tuesday August 06, 2024 @08:22AM (#41170168)
Your bar for interesting has to be insane then. What would you consider interesting if nothing from LLMs meets that bar?
Re:OpenAI co-founder John Schulman says he will leave and jo... by aleph_minus_one on Tuesday August 06, 2024 @09:11AM (#41170554)
For example there exist quite a lot of pure math papers that are so much deeper than basically every AI stuff that I have yet seen.
Re:OpenAI co-founder John Schulman says he will leave and jo... by snapcaster on Tuesday August 06, 2024 @11:05AM (#41171576)
So if LLMs weren't surprising to you, it would imply you expected this. If you did, how much money did you make on financial speculation? It seems like being this far ahead should have made you millions even without a lot of starting capital (look at NVDA alone)
Re:OpenAI co-founder John Schulman says he will leave and jo... by aleph_minus_one on Tuesday August 06, 2024 @11:52AM (#41172073)
> So if LLMs weren't surprising to you, it would imply you expected this.

I do claim that I have a tendency to be quite right about the "technological side" of such topics when I'm interested in them. On the other hand, events turn out to be different because of "psychological effects" (let me put it this way: I have a quite different "technology taste" than the market average).

In the concrete case of LLMs: the psychological effect why the market behaved so much differently is that I believed that people wouldn't fall for the marketing and hype of LLMs and would consider the excessive marketing to be simply dupery. The surprise to me was that this wasn't what happened.

Concerning NVidia: I believed that - considering the insane amount of money involved - people/companies would write new languages and compilers to run AI code on GPUs (or other ICs) of various different suppliers (in particular AMD and Intel) because it is a dangerous business practice to make yourself dependent on a single (GPU) supplier. Even serious reverse-engineering endeavours for doing this should have paid off considering the money involved. I was again wrong about this. So here the surprise was that lots of AI companies made themselves so dependent on NVidia.

Seeing lots of "unconventional" things is very helpful for doing math (often the observations that you see are the start of completely new theorems). Being good at stock trading and investing in my opinion on the other hand requires a lot of "street smartness".

Re:OpenAI co-founder John Schulman says he will leave and jo... by sanxiyn on Tuesday August 06, 2024 @04:08PM (#41174793)
Re: NVIDIA. I wholeheartedly agree. Google/TPU is an existence proof that it is entirely possible and rational to do so. My surprise was that everyone except Google missed.
Re:OpenAI co-founder John Schulman says he will leave and jo... by snapcaster on Wednesday August 07, 2024 @07:59AM (#41180470)
Okay so $0 it sounds like, you should figure out a way to monetize your future sight otherwise it comes off as cynicism masquerading as intelligence
Re:OpenAI co-founder John Schulman says he will leave and jo... by aleph_minus_one on Wednesday August 07, 2024 @12:02PM (#41182700)
> cynicism masquerading as intelligence

Rather: cynicism and a form of intelligence that is better suited to abstract math than investing. :-)

Re:OpenAI co-founder John Schulman says he will leave and jo... by claytongulick on Tuesday August 06, 2024 @10:53AM (#41171419)
I see it do a lot that's interesting but for programming stuff, I haven't found it to be particularly useful.

Maybe I'm doing it wrong?

I've been writing code for ~30 years, and I've built up patterns and snippets, etc... that are much faster for me to use than the LLMs.

A while ago, I thought I had a eureka moment with it when I had it generate some nodejs code for streaming a video file - it did all kinds of cool stuff, like implement offset headers and things I didn't know about.

I thought to myself, "self - you gotta check yourself, this thing is really useful".

But then I had to spend hours debugging & fixing the code that was broken in subtle ways. I ended up on google anyway learning all about it and rewrote everything it had generated.

For that case, while I did learn some interesting things from the code it generated, it didn't save me any time - it cost me time. I'd have learned the same things from reading an article or the docs on effective ways to stream video from the server, and I'd have written it more correctly the first go around.