The Cost of Over-Distillation: Why OpenAI Models Lost Their 'Taste'
zakelfassi · x · 2026-08-11
The author criticizes the new OpenAI model family for performing poorly in coding and conversation, calling its thinking trace 'slop.' They speculate that OpenAI relied too heavily on teacher-student distillation from a teacher model inherently lacking taste and judgment. In contrast, Anthropic's models (like Opus) rarely fail this way, leading the author to suggest that Anthropic hiring Instagram's co-founder was a move to instill taste. The core insight is that while intelligence distills perfectly, taste does not.
More from Models
- Claude Increases Lower Bound for Riemann Hypothesis to 67.2% — BoyNextDoor1990 · 2026-08-11
- Google's Gemini Omni Lite reportedly launching soon with focus on video — koltregaskes · 2026-08-11
- GPT 5.6 Sol Halves Coding Costs While Matching Previous Performance — jyangballin · 2026-08-11
- GPT 5.6 Sol Generates Novel Code Solutions, Not Just Memorization — jyangballin · 2026-08-11
- GPT 5.6 Sol Tops ProgramBench, Successfully Rebuilding Complex Programs — jyangballin · 2026-08-11
- Meta Rumored to Release Frontier Open-Weight Model Under Apache License — TimDarcet · 2026-08-11