Hello,
I am currently few days in the pro trial and have a question. I mostly use auto since in the pro trial the requests for premium models are gone pretty fast. However, I feel like sometimes the auto is performing extremely well, but then other times it suddenly starts to hallucinate a lot, says it has created features but didn’t even change a line of code, starts to randomly delete stuff etc. Is that something because of the trial? Or is the auto mode really something that may result in a very very bad LLM being selected?
Hey, this might be due to model hallucination. When the chat context gets overloaded, the model can start behaving unpredictably. To avoid this, try starting new chats regularly to keep the context clean.
I already do, I do a new chat for every new task but it still seems to really output trash ocasionally. It’s as if it puts me on a bad LLM some time and on a great other times and that more or less randomly. Is that accurate? And will it be exactly the same once the trial is over and I switch to a subscription? Or is the trial “worse” on purpose?