AI researcher: negative results matter and the pressure to splash is dialed to 11
code_star · x · 2026-11-05
An AI training researcher pushes back on the nihilistic tone around AI research: training models is "the best job in the world" — hard, confusing 90% of the time, but constantly surprising. Key points: negative results are as valuable as positive ones; establishing strong baselines beats running more experiments; be willing to change your mind when evidence arrives. He notes recent funding and attention have turned up the pressure for constant big wins to "11," but argues strong fundamentals always win because "the facts don't care about novelty." He closes urging team members to be kind and a source of joy, since everyone feels the pressure of this moment.
More from AGI Musings
- 'AI Is Like Electricity' Undersells It: AI Can Autonomously Move Atoms — danfaggella · 2026-11-06
- AP Exams Gutted: Biology Cut 40% of Content as Rigor Declines — garrytan · 2026-11-06
- Expedia CEO says AI agents aren't changing booking habits — yet — gaganghotra_ · 2026-11-06
- Both 'AI can't code' and 'AI solved coding' are mistakes — gdechichi · 2026-11-06
- Critics extend Dario's 'adults have more moral worth than babies' remark to billionaires — fkasummer · 2026-11-06
- "Word rotators" rise: AI shifts the valuable skill from spatial to verbal thinkers — StewartalsopIII · 2026-11-06