DeepMind Researcher Quips: RL 'Race to the Top' Boosts Capabilities, Not Safety
tomekkorbak · x · 2026-11-22
DeepMind researcher tomekkorbak quipped at Anthropic co-founder Eric Mitchell that the "race to the top" in reinforcement learning was supposed to lift safety standards across labs — instead, everyone seems to be racing to improve capabilities. A pointed jab at the safety-vs-capabilities tension in current RL competition.
More from Fun
- The death of a GPU: still rendering a world as its own world falls apart — snowclipsed · 2026-11-22
- AI asked to make a cartoon that survives deleting every frame puts a cat inside a .wav file — imjustnewatai · 2026-11-22
- Redditor makes an animated show starring Claude, coded by 20k+ lines of Claude-written Python — quaisar3 · 2026-11-22
- Gary Marcus clashes over neurosymbolic AI, claims Claude Code is neurosymbolic — mattyp · 2026-11-22
- Gary Marcus fires back: Claude Code is actually neurosymbolic AI — GaryMarcus · 2026-11-22
- Yudkowsky defines 'powerful AGI': biotech 10 years ahead of human state of the art — teortaxesTex · 2026-11-22