Yoav Goldberg blasts COLM mood: RL loss tweaks aren't 'science again'
yoavgo · x · 2026-10-11
AI researcher Yoav Goldberg calls out a pervasive sentiment at COLM: "last year was boring, we only did prompt tweaking, now we're doing science again." He argues that this 'science' mostly amounts to meaningless tweaks to RL loss terms, and he sees no reason it's any better than prompt tweaking — a sharp critique of how the field rebrands tinkering as progress.
More from AGI Musings
- Yoshua Bengio urges safety-minded AI employees to quit frontier labs: 'Stop pushing humanity to the brink' — Just-Grocery-2229 · 2026-10-11
- 10th grader used free Muse to produce 3 verified math preprints in 8 hours — EastConsequence3792 · 2026-10-11
- AI Impacts' famous 5% extinction risk stat was misleadingly framed, researcher says — jessi_cata · 2026-10-11
- Investor Ramez Naam: AI self-improvement loop is 5-10x too weak for takeoff — jessi_cata · 2026-10-11
- SRE veteran: AI won't replace human on-call engineers in my lifetime — RealGeneKim · 2026-10-11
- Karpathy: Think of LLMs as simulators, not entities with their own views — ZeroStateReflex · 2026-10-11