GPT3 is when the mass started to get exposed to this tech, it felt like a revolution.

Got 3.5 felt like things were improving super super fast and created that feeling the near feature will be unbelievable.

Got to 4/o series, it felt things had improved but users weren't as thrilled as with the leap to 3.5

You can call that bias, but clearly version 5 improvements displays an even greater slow down, that's 2 long years since gp4.

For context:

- gpt 3 got out in 2020

- gpt 3.5 in 2022

- gpt 4 in 2023

- gpt 4o and clique, 2024

After 3.5 things slowed down, in term of impact at least. Larger context window, multi-modality, mixture of experts, and more efficienc: all great, significant features, but all pale compared to the impact made by RLHF already 4 years ago.