A user has successfully trained a language model for DOS that fits on a floppy disk, utilizing Anthropic's Claude Fable 5.2 to write the model's transformer architecture and Claude Opus 5.5 to generate training data. The resulting model, with 367,000 parameters, can process and respond to approximately 3500 facts and conversational inputs, fitting within DOS memory constraints. AI
IMPACT Demonstrates the potential for AI models to generate and train smaller, specialized models for resource-constrained environments.
RANK_REASON User-driven research milestone involving AI models creating a functional, albeit limited, language model for an older operating system. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →