Guys could you advise me how to tell Ai to solve or add something without destroying more 3 things . Even when I share with him the whole code to make sure he revised it and give me accurate response I feel like he stop browsing the codebase when he find the first solution. and also it seems like he dont remember its previous messages or what it changed it.
give me some tips you use it guys to avoid this. Please!!
I see your pain. In my experience, the quality of all LLMs is not linear, there are good days and bad days. You need to be vigilant and check every answer. Long are the days you could rely on it, maybe because they are optimizing costs, but I would rather pay more to trust, than this human like behavior.
I lost at least 4 hours of work today with Claude Sonnet 3.5 which is the benchmark and used to be the best LLM for code ( I do python ) there wasn’t a single code change with 5 aditional prompts to fix the bugs created. And It was in average the best, but it’s a bad day for me.
My general tips :
- be vigilant, always check
- in doubt test the prompt + code section in a new chat with other LLM
- study and improve your prompting technique
- Posting to a forum reduces frustration and find more people with the same problem, it helps without solving
last one is very ironic, yet true and I simply can’t : - know what you can control and what you can’t . Don’t be upset with things you don’t control. This could require therapy, mindfullness, mental strenght, expensive and rare supplements and olympic training regimen.
1 Like
Yes you are right!
there is another problem that it may create new file in new folder even if it was exist so you must analyse not just its answer but also your project structure .
1 Like