About ten people were involved in finding a CAD structure problem that couldn't be found.
The model is still under research and cannot be shared externally. Since its validity expires next month and a new design will be released, I tried to get permission to cut out only the problematic area around it and throw it into Fable 5 to ask it to find the problem.
At the end of the max rate plan, I actually don't know. We just tried throwing it at AI once.
(I didn't try with a local llm, but I couldn't find it at the 100B model level.)
After burning about 70% of the max rate plan weekly limit in one go..(;ㅅ;) I finally found the cause.
It's easy to say, but the cause point presented in the analysis simulation code was inaccurate, and the cause was a 0.4 nm error in a pair of curves located about 2.5 cm away... How can humans find this...? ㅎㅎ
The last remaining Fable limit is to create a dataset generator that contains related problems and solutions.
Starting next week, I will train it using local llm and lora so that it can be solved.
I feel relieved now that the problem is solved, but I have a lot of worries about how to solve the same problem every time a new model is released... ㅎㅎ
