GPT5 is the most underhyped and impressive model I've seen

 Every single person I hear talking about GPT5 talks about how it was a 'flop' or 'overhyped'. How it wasn't world changing, how it didnt meet expectations, how it's not impressive, etc. 

However, before it came out, I had cancelled my OpenAI subscription, I renewed it for trying GPT5 and now I don't think I'll be cancelling it anytime soon. I had used o3, o4-mini and gpt 4.5

Before gpt5, gpt 4.5 was the most impressive model for me, in terms of actually being something close to semi decent writer and actually having some useful information. 

By and large, the vast majority of models have been absolutely trash at research for me, for anything expect for very general or basic stuff in fields. Anything that required actual reasoning ability - the ability to actually put things together and form new ideas from them, they would just fail at. 

e.g. taking ideas from neuromorality and reinforcement learning and using that to put together a new reinforcement learning method that uses ideas from neuromorality taking ideas from how info in emotional situations seems to reach the amygdala first, then the prefrontal cortex, to make a similar small specialized model that gets info first and a larger model that gets info from it. 

Try to do this with Claude opus 4.1, gemini 2.5 pro, etc, and they have some facts in a list, essentially, but gpt5 thinking hard is the first model ever where i can see some actually new ideas being made.

not huge. but very very new.

Comments

Popular posts from this blog

Pain of Writing

we should surpass ai

Truth