Model specification (artificial intelligence)
A model specification is a document that specifies the intended behavior of a large language model (LLM).[1][2][3][4][5][6] The specification may include core principles or prohibitions intended to prevent undesired behavior as part of AI alignment.[6][3] The company Anthropic includes their model specification as part of their concept of "constitutional AI".[7][8][9][10][11][12]
The EU AI Act's General-Purpose AI Code of Practice requires signatories to provide a model specification.[13][14][15][16]
Research has been done into the effectiveness of model specifications.[17][18]
Dean Ball and Daniel Kokotajlo advocated in an article published by Time magazine that companies should be required by regulation or industry standards to publish their model specifications publicly.[19]
See also
- Constitutional AI
- AI alignment
- AI safety
- Reinforcement learning from human feedback
- Artificial Intelligence Act
- Three Laws of Robotics
References
- ↑ "EU AI Act: General-Purpose AI Code of Practice". EU AI Office. Retrieved March 21, 2026.
- ↑ "Introducing the Model Spec". OpenAI. May 8, 2024. Retrieved March 21, 2026.
- ↑ 3.0 3.1 "Claude's Constitution". Anthropic. May 9, 2023. Retrieved March 21, 2026.
- ↑ "Sharing the latest Model Spec". OpenAI. February 12, 2025. Retrieved March 21, 2026.
- ↑ "openai/model_spec". GitHub. Retrieved March 21, 2026.
- ↑ 6.0 6.1 "Model Spec (2025/12/18)". OpenAI. Retrieved March 21, 2026.
- ↑ Perrigo, Billy (January 21, 2026). "Anthropic Publishes Claude AI's New Constitution". TIME. Archived from the original on January 24, 2026. Retrieved March 21, 2026.
- ↑ "AI startup Anthropic wants to write a new constitution for safe AI". The Verge. May 9, 2023. Retrieved March 21, 2026.
- ↑ "Claude's new constitution". Anthropic. January 22, 2026. Retrieved March 21, 2026.
- ↑ "Anthropic writes 23,000-word 'constitution' for Claude". The Register. January 22, 2026. Retrieved March 21, 2026.
- ↑ Ropek, Lucas (January 21, 2026). "Anthropic revises Claude's 'Constitution,' and hints at chatbot consciousness". TechCrunch. Retrieved March 21, 2026.
- ↑ "Interpreting Claude's Constitution". Lawfare. January 21, 2026. Retrieved March 21, 2026.
- ↑ "The General-Purpose AI Code of Practice". European Commission. Retrieved March 21, 2026.
- ↑ "The EU's General Purpose AI Code of Practice: What You Need to Know". Deloitte. Retrieved March 21, 2026.
- ↑ "EU's General-Purpose AI Obligations Are Now in Force, With New Guidance". Skadden, Arps, Slate, Meagher & Flom. August 2025. Retrieved March 21, 2026.
- ↑ "Article 99: Penalties | EU Artificial Intelligence Act". Retrieved March 22, 2026.
- ↑ "Stress-testing model specs reveals character differences among language models". Anthropic. Retrieved March 21, 2026.
- ↑ "Collective Constitutional AI: Aligning a Language Model with Public Input". Anthropic. Retrieved March 21, 2026.
- ↑ Ball, Dean W.; Kokotajlo, Daniel (October 15, 2024). "4 Ways to Advance Transparency in Frontier AI Development". TIME. Retrieved March 21, 2026.
This article "Model specification (artificial intelligence)" is from Wikipedia. The list of its authors can be seen in its historical and/or the page Edithistory:Model specification (artificial intelligence). Articles copied from Draft Namespace on Wikipedia could be seen on the Draft Namespace of Wikipedia and not main one.
