ML pro tip: make ugly training curves look great with log-log axes and max smoothing
willdepue · x · 2026-11-21
Will Depue shares a simple ML trick: if your training charts look bad, just switch both axes to log scale and crank smoothing to max — "it never fails." He quips with a Fritz Houtermans line: "even my grandmother looks like a straight line on a log log plot." The attached screenshots show a training-loss dashboard where per-head and per-data-source loss curves all look smooth and elegant under log-log axes with maximum smoothing.
More from Fun
- AI meme: 'Anthropic announces fast takeoff as probes exit the heliosphere' while $ANT drops 7% — aiamblichus · 2026-11-21
- Ramayana epic recreated as a GTA-style boss battle with AI video and dubbing — umesh_ai · 2026-11-21
- Blogger shows how to max out Gemini Ultra perks: 'I guess I own an H100 now' — yihui_indie · 2026-11-21
- Joke: Aschenbrenner means 'burns things to ashes' in German — nominative determinism strikes — minhnguyen · 2026-11-21
- Jensen Huang repeatedly interrupted Mistral CEO at panel: "You're not even speaking human" — mattturck · 2026-11-21
- ERC reviewer rejects LLM autoformalization proposal: "tried without good results" — marc_lelarge · 2026-11-21