Screenshot of this question was making the rounds last week. But this article covers testing against all the well-known models out there.
Also includes outtakes on the ‘reasoning’ models.
Screenshot of this question was making the rounds last week. But this article covers testing against all the well-known models out there.
Also includes outtakes on the ‘reasoning’ models.
I think its far more likely to degrade itself in a feedback loop.
It’s already happening. GPT 5.2 is noticeably worse than previous versions.
It’s called model collapse.
To clarify : model collapse is a hypothetical phenomenon that has only been observed in toy models under extreme circumstances. This is not related in any way to what is happening at OpenAI.
OpenAI made a bunch of choices in their product design which basically boil down to “what if we used a cheaper, dumber model to reply to you once in a while”.
The funny thing is, in order to get it to the dumber model, they have to run people’s queries through a model that selects the appropriate model first. This is resulted in new headaches for AI fans
Yeah that’s also something that you have to train for, i’m not super aware of the technicals but model routing is definitely important to the AI companies. I suspect that’s part of why they can pretend that “inference is profitable” as they are already trying to squeeze it down as much as possible.
I wonder if the routing is actually going to decrease the overall costs or increase them… Routing looks like it introduces new, unavoidable factors that would cause the costs to increase.
Yeah i remember that Ed article ! I don’t think the technical aspects are relevant to the newer generation of models, but yeah of course any attempt to compress inference costs can have side effects : either response quality will degrade for using dumber models, or you’ll have re-inference costs when the dumb model shits its pants. In fact the re-inference can become super costly as dumber models tend to get lost in reasoning loops more easily.
I mean, we’re watching it happen. I don’t think it’s hypothetical anymore.
I’m sorry but no, models are definitely not collapsing. They still have a million issues and are subject to a variety of local optima, but they are not collapsing in any way. It is not known whether this can even happen in large models, and if it can it would require months of active effort to generate the toxic data and fine-tune models on that data. Nobody is gonna spend that kind of money to shoot themselves in the foot.
Then why are newer versions of the major models performing so poorly? For instance, GPT 5.2 is definitely not an improvement over 4.5. What’s the root cause?
I feel that a lot of what is improving in the recent batch of model releases is the vetting of their training data - basically the opposite of model collapse.
Nothing requires an LLM to train on the entire internet.
That’s an excellent point! On that topic I recently listened to an interview of the founder of EleutherAI, who focuses on training small language models. She said they were able to train a 1B parameters reasoning model with 50K Wikipedia articles and carefully curated RL traces. The thing could run in your smartphone and is at parity with much larger models trained on trillions of tokens.
She also scoffed at Common Crawl and said it contained mostly cookies and porn. She had a kind of attitude like “no wonder the big labs need to slurp trillions of tokens when the tokens are such low quality”. Very interesting approach, if you understand french I can only recommend the interview.
The very interesting part will be how successful they are at training the training data selectors to choose high quality data sources.
I think a lot of it is still done by hand, and there is also synthetic data distilled from larger models of course.