What's this about?
A weekend debate about *how frontier models write*. Their dense, jargon-packed prose style — nicknamed "Fableish" (after Anthropic's Fable) or "Sol-speak" (after GPT-5.6 Sol) — is defended by some as evidence the models are just smarter than us. Others, including Ethan Mollick, argue it's the opposite: it's a failure of the model to model *you*, the reader. Efficient-for-the-model output is not the same as clear-for-the-human output.
What is "theory of mind"?
Theory of mind is the ability to model what someone else knows, needs, and can understand — and to adjust your communication accordingly. A good teacher has it; they explain differently to a beginner than to an expert. The critique is that when a model dumps maximally-compressed jargon, it's writing for itself, not for the person reading — a theory-of-mind gap. "Most efficient for Fable" can still be "a failure applied to the user."
What's the community saying?
- Smart vs. lazy — one camp reads the dense style as high intelligence; the other as the model optimizing the wrong objective (its own token efficiency, not reader comprehension).
- Product angle — Mollick extends it: ChatGPT Work and Claude Cowork fail non-coders because they assume non-engineers can't reason about work the way engineers do.
- Can you strip it out? — echoes last week's thread about "de-AI-ifying" AI writing: people want the substance without the tells, but the style keeps resurfacing.