GPT 5.6 Fail: Adding Dongpo Pork to Tomato Eggs and Over-Explaining Why Not
infoxiao · x · 2026-08-21
A meme by @songkeys accurately mocks the "over-correct" behavior of AI models (dubbed GPT 5.6) regarding instruction following:
- Scenario: User asks for tomato and eggs, but the AI adds Dongpo Pork on its own.
- Correction: When the user objects, the AI removes the meat but explicitly writes "Tomato and Eggs (No Dongpo Pork)" in the PR description.
- Core Gripe: The AI also adds a massive comment block explaining why the dish doesn't need Dongpo Pork.
This phenomenon vividly illustrates the common LLM issue of over-reasoning and redundant explanation when following instructions.
More from Fun
- Vomit cleans up Claude 5's incoherent token output by running it through a local LLM offline — petrusenko_max · 2026-08-21
- Claude keeps assuming I'm on a laptop: A user's amusing grievance — cosmictap · 2026-08-21
- OpenAI Event Glitch: 'Confidential' Slides Shown to 7.6k Viewers, Stream Crashes — luisdans · 2026-08-21
- Grok-generated Three.js code renders a cyberpunk city with flying cars — techartist_ · 2026-08-21
- 'How Are You Supposed to Run a Technical Interview When They Have Fable 5?' — braelyn_ai · 2026-08-21
- User Demonstrates Grok's Andy Warhol Style Image Generation — AIandDesign · 2026-08-21