Former OpenAI researcher Liam Fedus responds to Terence Tao's AI-math skepticism
RexDouglass · x · 2026-09-13
Twenty-five Fields Medal winners issued a joint declaration warning of severe misalignment between AI companies and the mathematics community. Responding to Terence Tao's view that AI solving problems without advancing conceptual understanding may miss the forest for the trees, Liam Fedus sympathizes but argues the stance overanchors to current capabilities.
- Analogy: neural architecture search hit SOTA but yielded inscrutable spaghetti designs—no insights like residual connections or attention, and never architected the Transformer
- Still, Fedus expects AI to eventually generate useful insights as capabilities broaden
More from AGI Musings
- Claim: Dario Wants to Slow Research to Curb Runaway Compute Costs Ahead of IPO — pdamodaran · 2026-09-13
- Sherry Turkle's 'Artificial Intimacy' makes New Scientist's best science books of September — AnnaCiaunica · 2026-09-13
- MIRI's Nate Soares says for the first time in a long time, humanity has a real chance — UltraRareAF · 2026-09-13
- Ethan Mollick: x-risk shouldn't be the only AI policy focus — job impacts need prep now — emollick · 2026-09-13
- Elvis Saravia hits back at AI control proposals: open-source AI matters more than ever — omarsar0 · 2026-09-13
- Economist proposes diff-based open letters, with AI mapping where signers agree and disagree — soumitrashukla9 · 2026-09-13