I think AI companies might be intentionally configuring their smart models to not solve problems quickly
I have a bold hypothesis.
I think AI companies might be intentionally configuring their smart models to not solve problems quickly, particularly the coding requests, because if they did, their revenue would be severely impacted by quick and short solutions.
If the model over engineers, then I need to talk to it a lot to fix the problem, particularly with setups that use APIs, so they get way more tokens burnt. SO more revenue for them (or I need a more unlimited subscription).
Yea, I know, sounds insane. But think about your experience with code generation — wasn’t it significantly more efficient before the models became smart/thinking/whatever? How come these significantly better models produce significantly more over engineering, where the only benefit is burning tokens?