>>105796022 (OP) Have the interface ask the model your question 100 times, then challenge the model with a contradictory statement, then fuse the results if all those responses, have a model identify references, then have another 100 models assess the accuracy of the result taking the references as context, then fuse those assessments and have a model generate a final response
>>105796022 (OP) LLMs don't "know" anything, they are incapable of reason. They can only pretend to know things and reason, by being good at the one thing they can do, which is guess the next token. Over and over.
I gave jeetGPT a txt file with terminal log of error messages and asked it identify the problem based on it. It immidiately told me the problem was X and quoted certain passages from the file.
Went into the file and tried looking for them and they werent there at all. Confronted jeetGPT and it admittet it just made it up because it thought that was the likely problem to what I had described and it was faster than actually reading the txt file.
>>105796239 >Went into the file and tried looking for them and they werent there at all. Confronted jeetGPT and it admittet it just made it up because it thought that was the likely problem to what I had described and it was faster than actually reading the txt file. kek >design machine in the image of the human brain >it emulates human laziness and dishonesty
Anonymous
7/4/2025, 9:14:18 AM No.105796668
>>105796224 "You pencil pusher"? Is that what he's saying?
Anonymous
7/4/2025, 12:12:52 PM No.105797742
>>105796239 I like gemini as a talking rubber duck. I like asking it what could be the cause and giving me a list of starting points to diagnose from. Then I look at whatever I think is good and figure it out from there.