[R] debugging-only LLM? chronos-1 paper claims 4–5x better results than GPT-4 ... thoughts? Retro

i stumbled on a paper about a model called chronos-1 that’s trained purely on debugging workflows ... no autocomplete, no codegen, just stack traces, logs,...