Post-training pipelines remain highly manual and far from automatable
menhguin · x · 2026-11-06
In a debate on whether post-training is verifiable/automatable, menhguin argues current post-training data pipelines are extremely manual and human-directed — its whole point is pushing model capability distributions toward economically valuable skills. Inference engineering is automatable, pretraining somewhat, but post-training is far from it today.
Related event: Post-training pipelines remain heavily manual, AI researchers say(2 posts)→
More from AGI Musings
- alvelda proposes embedding agent memories into model weights, like sleep consolidates memories — alvelda · 2026-11-06
- AI circle debate: can artists who shun AI still be called great artists? — repligate · 2026-11-06
- Enterprise AI: Judgment, Not Intelligence, Is the Scarce Resource — AI Engineer · 2026-11-06
- Bank of England report: tech growth driven by company spending, not worker efficiency — GaryMarcus · 2026-11-06
- Schmidhuber: Supersmart AIs will transform the galaxy in a few hundred thousand years — SchmidhuberAI · 2026-11-06
- Musk: civilizations follow a life cycle like organisms, none lasts forever — XFreeze · 2026-11-06