LLMs / 2026-09-17

Note: An LLM was not used in writing this microblog entry.

This entry is part of a microblog series called LLMs

---

Not a thought entry, more of a data and notional thoughts for reformulation later.

News:

• Mathematicians are having their AI disillusionment moment, Terrence Tao leads, Lisp Curse overlap? See also LLM Curse.
• Martin Fowler is also having his doubts, but the reasoning is immature, lacking any insight. Odd.
• Dan Luu seems to be getting in on the action too, criticising the meat proxy, on, as always, quantitative terms rather than qualitative terms.

Germinating thoughts:

• Boilerplate redefined as that which can be produced without much thought, and therefore all LLM output is inherently boilerplate. Discuss implications against historical programming conventional wisdom with regards to boilerplate.
  • Hint: the ratio of prompt to code could be a measure of language power. The Next 700 Programming Languages - perhaps different?
• Barefoot shoes vs “supportive” shoes: the supportive ones alter the wearer until the wearer can’t walk unassisted.

Links:

• Solomonoff’s theory of inductive inference
• No one actually wants simplicity
• Hydrating pizza
• Claude is your insider threat now - Dan Tentler - Security Fest 2026 - YouTube — good grief.