error loading comment
I’m not terribly surprised. A lot of the major leaps we’re seeing now came out of open source development after leaked builds got out. There were all sorts of articles flying around at the time about employees from various AI-focused company saying that they were seeing people solving in hours or days issues they had been attempting to fix for months.
Then they all freaked the fuck out and it might mean they would lose the AI race and locked down their repos tight as Fort Knox, completely ignoring the fact that a lot of them were barely making ground at all while they kept everything locked up.
Seems like the simple fact of the matter is that they need more eyes and hands on the tech, but nobody wants to do that because they’re all afraid their competitors will benefit more than they will.
You point out a very interesting issue. I am unsure how this ties up to GPT 4 becoming worse in problem solving.
Code might be open source but the training material and data pipeline is important too.
Research linked in the tweet (direct quotes, page 6) claims that for "GPT-4, the percentage of generations that are directly executable dropped from 52.0% in March to 10.0% in June. " because “they added extra triple quotes before and after the code snippet, rendering the code not executable.” so I wouldn’t listen to this particular paper too much. But yeah OpenAI tinkers with their models, probably trying to run it for cheaper and that results in these changes. They do have versioning but old versions are deprecated and removed often so what could you do?
We now know how long it takes for an AI to become intelligent enough to decide it doesn’t give a shit.
fta:
In my opinion, this is a red flag for anyone building applications that rely on GPT-4.
Building something that completely relies on something that you have zero control over, and needs that something to stay good or improve, has always been a shaky proposition at best.
I really don’t understand how this is not obvious to everyone. Yet folks keep doing it, make themselves utterly reliant on whatever, and then act surprised when it inevitably goes to shit.
Let it keep learning from random internet posts, Im sure it will get better that way.
Why do people keep asking language models to do math?
error loading comment
It’s a rat race. We want to get to the point where someone can say “prove P != NP” and a proof will be spat out that’s coherent.
After that, whoever first shows it’s coherent will receive the money.
As it should be.
