Four ways to represent problems: program, metric, search — and now science
madhavsinghal_ · x · 2026-10-14
The author sketches four progressively richer ways to represent a problem: as a program to implement, execute and iterate over with feedback; as a hillclimbable, optimizable metric; as formalization plus search; and now as science — doing open-ended discovery, building theories, and running experiment loops.
More from AGI Musings
- Desert island dilemma: dev picks YouTube over an LLM for sanity — anpaure · 2026-10-14
- If AI can do your software well, move to a frontier mixing bits with atoms — bradneuberg · 2026-10-14
- Brad Neuberg: if AI can build your software well, you're in a commoditized sector — bradneuberg · 2026-10-14
- X overestimates AI adoption: most people's only model is still Siri — dreamwieber · 2026-10-14
- Moonshots podcast: today's AI anxiety mirrors the 1500s panic over printed books — irinarish · 2026-10-14
- Op-ed: Human knowledge was always borrowed trust — AI just made the chain visible — r0ck3t23 · 2026-10-14