DSPy Creator: Your Moat Is Not Finetuning vs Prompting, It's the Spec
lateinteraction · x · 2026-08-14
DSPy creator @lateinteraction responds to Paul Graham's tweet about the return of tuning open-weight models, noting that DSPy has been right about this since Feb 2023:
- Finetuning, prompting, and scaffolding are all automatable implementations of the same spec.
- Your moat shouldn't be an implementation detail.
Paul Graham's original tweet: A year ago lots of startups were tuning open weight models. Then the consensus became that this was a waste of time, because the model companies' next versions would surpass them a couple months later. Now tuning open weight models seems to be back.
More from AGI Musings
- AI and the Ownership of Judgment: Aspen Seminar Reading List Sparks Reflection — chalmermagne · 2026-08-14
- AI agent safety protocol: user accountability, infrastructure-enforced constraints — AsleepSwordfish9603 · 2026-08-14
- Mathematician Criticizes Turning Open Problems into AI Benchmarks — rbhar90 · 2026-08-14
- Peter Diamandis: Your Child Is Not Ready, Education Needs Reform for AI Era — PeterDiamandis · 2026-08-14
- Terence Tao: AI Will Transform Math, Forcing Reflection on Its Goals and Values — AlexKontorovich · 2026-08-14
- Sam Altman: Only 50 people once believed AGI possible, 45 of them at OpenAI — garrytan · 2026-08-14