Not sure I'm more productive with LLMs. They keep making up names of programs and arguments and things, so I need to try, I get an error, go back to the prompt and it admits that I was right, so it suggests a second thing... until eventually it works, maybe. Seems like a waste of time. I tried finding an answer by searching the web, found a blog post written by a human and that just worked.
Using Venice and Perplexity. Maybe there's something better?
Yes, llms with bigger context window and rag can be feed with all the documentation of whatever software and be much more accurate in problem solving tasks
Makes sense, which one do you recommend?
I’ve use all the major ones quite often and unfortunately they’re still not ready for prime time. Error rates and hallucinogens are just too high.
Forgot to mention, in my experience, Perplexity is probably still the best of the bunch.
I feel the same. I have however found better usage, tedeus text formatting. For example giving a completely messed up text from web and returning a markdown table works well. Actual cli commands fail pretty quickly and exponentially with more number of arguments.
Replit is fantastic. Finally has Claude support as well.
