"Never Attribute to Malice What Can Be Explained by Bad Post-Training," AI Researcher Quips
lfschiavo · x · 2026-09-29
X user lfschiavo riffed on the classic Hanlon's razor: "don't ascribe to malice what can better be ascribed to bad post-training." The quip suggests most problematic model behavior stems from training-quality issues rather than intent, and it resonated across the AI community.
More from Fun
- Even AI insiders got fooled: an AI-generated person they swore was real — bennash · 2026-09-29
- Sending Tens of Thousands to ID-Less X Money Accounts Ends Exactly As Expected — nptacek · 2026-09-29
- Late? You're 'pacing the frontier': AI jargon as life's universal excuse — SuB8u · 2026-09-29
- How do you make a chess bot blunder believably? Mixing Maia and Stockfish isn't enough — space64-llc · 2026-09-29
- Visitor meets OpenAI's Codex lead Thibault Sottiaux, confirms the 'physical reset button' — DeryaTR_ · 2026-09-29
- Real-time Among Us demo pairs DeepSeek V4 Flash planner with Jev actor agent — ai · 2026-09-29