Fine-Tuned Open Source Models Cost 95% Less and Train in Under 48 Hours, Challenging Frontier Labs' Moat
ayushtweetshere · x · 2026-09-14
A quoted thread argues that fine-tuned open-source models trained on custom data cost about 95% less than frontier models, perform better, and can be trained in under 48 hours. The implication: big AI labs have no moat even at the enterprise level, since any rational company would prefer a custom open-source model over paying API pricing to Anthropic and OpenAI. If open source keeps growing, frontier labs' business collapses — leaving FUD and regulation as their only play. The reposter adds that high-EQ AGI models were sunset in favor of "lobotomized substitutes" and declares China the winner.
More from AGI Musings
- Rob LeClerc: quoting p(doom) without conditional probabilities reveals shallow thinking — robleclerc · 2026-09-14
- Proposal: an 'aiXiv' venue for AI-written, verified research with no human author — tdietterich · 2026-09-14
- AI-written papers flood arXiv: Dietterich proposes oral exams and an aiXiv venue for AI research — tdietterich · 2026-09-14
- AI Regulation Needs an Asilomar-NPT Playbook, Not a 'China Will Win' Test — krishnan · 2026-09-14
- Math training makes you a stronger LLM user, even beyond math topics — josh_wills · 2026-09-14
- Ethan Mollick: People who never cared about AI are now freaking out about it killing everyone — emollick · 2026-09-14