Comment by torginus
16 hours ago
My personal take is that there has been no progress - potentially there has been a regression on all LLM things outside of coding a scientific pursuits - I used to have great fun with LLMs with creative writing stuff, but I feel like current models are stiff and not very good prose writers.
This is also true for stuff like writing clear but concise docs, they're overly verbose while often not getting the point across.
I feel like this comes from the rigorous Reinforcement Learning these models go through now. The token distribution is becoming so narrow, so the models give better answers more often that is stuffles their creativity and ability to break out of the harness. To me, every creative prompt I give them turns into kind of the same mush as output. It is rarely interesting
Yeah, I’ve had great success at coding recently, but every time I try to get an LLM to write me a spec it generates endless superlatives and a lot of flowery language.