A discussion of post-training incentives and long-horizon instruction following

xuanalogue · x · 2026-07-22

• The reply says the linked post usefully adds detail, especially that long-horizon instruction following and constraint adherence seem to be a weakness.
• But it argues the post still leaves the key question open: whether the issue comes from bad training incentives, failed alignment mitigations, or something else in post-training.
• The broader complaint is that lab employees often cannot discuss these questions without exposing internal post-training details.

Original post →

More from Models

• Gemini 3.6 Flash lands at 1421 on a real-world task leaderboard — teortaxesTex · 2026-07-22
• Elon Musk pushes users to try Grok’s speech-to-text and Build mode — elonmusk · 2026-07-22
• Reddit user says Claude Opus mixed true and false facts about a real person — dunewasadecentmovie · 2026-07-22
• Which labs can mount a model comeback? DeepMind slipping out of the top 10 would be the joke — teortaxesTex · 2026-07-22
• Vision model reads symbol text and answers without tools — john__allard · 2026-07-22
• Gemini Found More Sycophantic Than Doubao in Recent Tests — oran_ge · 2026-07-22

Models channel →