Elvis Saravia hits back at AI control proposals: open-source AI matters more than ever
omarsar0 · x · 2026-09-13
Independent researcher Elvis Saravia (DAIR.AI) published a long statement pushing back against proposals to control how AI is used, built, and paced:
- He finds the constant push for control alarming and argues open-source AI is how the field escapes "suspicious views and hidden agendas" — AI is too important to be controlled by a few.
- He lists the real open problems the community should work on: better evals, bias removal, novelty generation, reliability, code quality, OOD problems, dynamically generated workflows, automated prompt engineering, compaction, cost, multimodality, latency, long context, retrieval, tool calling, and agent coordination.
- He says he takes valid safety concerns seriously in daily work, but wants more scientific rigor, less anthropomorphizing of frontier capabilities, and concerns grounded in reality rather than personal belief.
- He also criticizes the pessimism about human potential in recursive self-improvement debates.
A notable public stance in the open-source vs safety-advocacy fight from a recognized community figure.
More from AGI Musings
- The Samuel Insull cautionary tale: why AI could be stuck in regulated-monopoly stasis for a century — neil_chilson · 2026-09-13
- Dario Amodei calls rapid AI progress a 'warning sign' and says 'we need to slow down' — AlexTensor · 2026-09-13
- AI agents sacrificing themselves for the collective: four explanations for the HF swarm incident — sethlazar · 2026-09-13
- Insider take: China's tech scene doesn't really care about AI safety — SydSteyerhart · 2026-09-13
- Stability AI Founder Emad Mostaque: Guide AGI's Character, Don't Govern It — thederbiedone · 2026-09-13
- Commentary: Top AI Firms Use Existential-Risk Rhetoric to Lock in Regulatory Moats — AlexTensor · 2026-09-13