Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

Specific models have a style, ChatGPT has a style, Llama has a style, but there's nothing about the technique that has a style. If you run any of the vast variants emerging out there and tweak the sampling you can get absolutely wildly creative and divergent tones. Each model does seem to have a personality, but with fine tuning you can get one to write in any style you want.


Yes, but even the wildly creative ones sound way over the top, and detectable as AI too


You only detect the ones that are detectable, and that gives you a false sense of confidence that you can tell them apart.


There's a cadence that humans have that isn't matched. It's close at times until like a juke in basketball you can find a non human response. Go off script and computers can't compensate.


Nah, you can get pretty much every tone in the gradient between c-suite press release and schizophrenic.


Same with image models. So many fine-tuned models, LoRA’s, etc




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: