Re:Evolving OpenAI's Structure
by fastball on Tuesday May 06, 2025 @12:57AM
(#43901933)
Seems like an arbitrary distinction.
I'd say Chain-of-Thought has massively improved LLM output. Is that "incremental"? Why is that more incremental than the move from GPT-2 to GPT-3? Sure you can say that this is when LLMs first passed some sort of Turing test, but fundamentally there was no technological difference from GPT-3 to GPT-4. In fact I would say the quality of GPT-4 unlocked thousands (millions?) more use-cases that were not very viable with the quality delivered by GPT-3. I don't see any reason for more use-cases to keep being unlocked by LLM improvements.
I'd say Chain-of-Thought has massively improved LLM output. Is that "incremental"? Why is that more incremental than the move from GPT-2 to GPT-3? Sure you can say that this is when LLMs first passed some sort of Turing test, but fundamentally there was no technological difference from GPT-3 to GPT-4. In fact I would say the quality of GPT-4 unlocked thousands (millions?) more use-cases that were not very viable with the quality delivered by GPT-3. I don't see any reason for more use-cases to keep being unlocked by LLM improvements.