What Should Go In A Model Spec?
forethought.org · 3,726 words · saved by 1 readers
Forethought considers candidate criteria for deciding what should go in an AI's model spec.
4th June 2026 [AI Narration] What should go in a model spec? Playback speed Volume 0:00 of 22:33 Suppose an AI company is considering whether to include some particular quality X – a rule, virtue, heuristic, default, attitude, goal, or style – in a model spec. Perhaps they are considering whether their LLM should have prosocial drives. Perhaps they’re wondering if the LLM should whistleblow to help prevent extreme power concentration. Or perhaps they’re uneasy about whether the LLM should be so exactingly honest that it always tells the truth to children about Santa. And so on. What…
saved by
related reading
- Stickiness in AI Behavioral Designforethought.org
- The Persona Selection Model: Why AI Assistants might Behave like Humansalignment.anthropic.com
- The behavioral selection model for predicting AI motivations — LessWronglesswrong.com
- Human Judgment as a Specificationblog.brownplt.org
- Introducing the Model Spec | OpenAIopenai.com
- Model Spec Midtraining: Improving How Alignment Training Generalizesalignment.anthropic.com
- Patterns for Building LLM-based Systems & Productseugeneyan.com
- Should We Train Against (CoT) Monitors? — LessWronglesswrong.com
- Taking LLMs Seriously (As Language Models) — LessWronglesswrong.com
- Model Spec (2024/05/08)cdn.openai.com
- Stress-testing model specs reveals character differences among language modelsalignment.anthropic.com
- Self-Fulfilling Misalignment Data Might Be Poisoning Our AI Modelsturntrout.com