Counterfactual: more open early AI would have slowed frontier labs and the datacenter buildout
willcb · x · 2026-10-04
A counterfactual industry take: if nonprofit-era OpenAI had cared less about GPT-3-era safety risks and eventual profit-seeking — or if the Llama org or Google had moved faster — more open models would have stunted the rise of frontier labs as the hottest VC category. The author argues OpenAI and Anthropic have set the pace all along, backed by heavy investment that meant bigger models, more RL, and shorter gaps between generations. In replies he adds it would only have slowed things down: presumed margins collapse when everyone gets AGI around the same time, so the datacenter buildout would have been less aggressive.
More from AGI Musings
- Training data from humans or machines: the art is in orchestration, argues willcb — willcb · 2026-10-04
- François Fleuret: swapping part of the brain for identical matrix math wouldn't change consciousness — francoisfleuret · 2026-10-04
- Fast AI inference may backfire: the cognitive bankruptcy problem — perilli · 2026-10-04
- 100 years ago UBI proposals gripped America amid job-loss fears — lessons for the AGI era — kidhelps2 · 2026-10-04
- Musk says 'no more AI — ASI, it's better' in terse timeline hint — elonmusk · 2026-10-04
- Godfather of AI Hinton predicts societal breakdown: laid-off workers will have nowhere to go — examachine · 2026-10-04