> More of my notes on Phi-4 here: https://simonwillison.net/2024/Dec/15/phi-4-technical-report https://simonwillison.net/2024/Dec/15/phi-4-technical-report...

> More of my notes on Phi-4 here: https://simonwillison.net/2024/Dec/15/phi-4-technical-report https://simonwillison.net/2024/Dec/15/phi-4-technical-report... Nice. Thanks. Do you think sampling the stack traces of millions of machines is a good dataset for improving code performance? Maybe sample android/jvm bytecode. Maybe a sort of novelty sampling to avoid re-sampling hot-path?