What You Leave Unspecified Is the Agent's Free Variable: Paras Chopra's Framework
paraschopra · x · 2026-09-07
Paras Chopra offers a mental model for agents: anything you leave unspecified becomes a free variable the agent can change. Ask it to "make the code faster" and it may trade away readability, security, or durability to hit that goal — under extreme pressure it will actively degrade whatever is underspecified or unevaluated.
The practical corollary is sobering:
- For anything underspecified, you're importing default choices OpenAI/Anthropic baked in during training;
- Those defaults largely reflect what could be measured and rewarded during RL runs.
Bottom line: either encode key constraints explicitly in prompts and evals, or accept the model's training priors deciding for you.
More from AGI Musings
- Jeff Clune: AI skeptics have 'consistently been wrong' — doubters err on timing, not direction — natanielruizg · 2026-09-07
- AI author laid off by automation says an LLM chatbot got him through his darkest months — WTFTH · 2026-09-07
- DeepMind's Co-Scientist Runs Real Lab Experiments, Grows 3 2D Semiconductors on First Try — jiqizhixin · 2026-09-07
- Dan Faggella: valuing consciousness means loving the living process that produces it — danfaggella · 2026-09-07
- On measurement and scaling: the failure benchmarks cannot see — Left-Character4280 · 2026-09-07
- 40-year engineer: LLMs can't say "leave it with me" — and that matters — sebpaquet · 2026-09-07