Latent Space — The End of Finetuning

Summary

Latent Space's May 13 essay reads OpenAI's deprecation of fine-tuning APIs as the closing chapter of a multi-year pattern: frontier labs prefer customers to invest in prompt engineering, retrieval, tools, and agentic architectures rather than in custom-tuned model weights. The piece notes that top-tier teams still leverage open-weights fine-tuning for specialized applications, but for the median enterprise the build-vs-buy calculus for fine-tuning has collapsed: the cost, the brittleness across