Tim O'Reilly: Big AI Labs Are Building the Wrong AI
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Tim O'Reilly argues big AI labs are misreading the future by optimizing for the biggest frontier models rather than the use cases ordinary people want.
- O'Reilly distinguishes true open-source AI — a full stack with clean separation between model, harness, and application — from merely open-weight models, saying current labs have built an 'architecture of control' over users.
- O'Reilly counters the security argument against open-source AI by noting that every cybersecurity incident so far has come from frontier models, making them the case for slowing frontier development, not restricting open weights.
- O'Reilly warns the US could 'win' frontier AI and still lose to China, because Beijing is diffusing lower-level models widely through society while American labs chase capability.
- O'Reilly's nonprofit AI Disclosures Project is developing an 'open-memory consortium' that would let users switch models and providers while preserving context, directly countering Mark Zuckerberg's Meta lock-in thesis.
- O'Reilly acknowledges AI is disrupting his own business, noting O'Reilly Media's book revenue has fallen from a $70 million peak to roughly $30 million.
- O'Reilly calls Silicon Valley 'anti-capitalist' since roughly 2010, pointing to VC-subsidized ride-hailing giants like Uber and Lyft as the template that replaced market-driven outcomes.
Why it matters: O'Reilly's critique carries weight because he helped define the 1990s open-source era and now sees labs repeating Microsoft's 1990s lock-in playbook. His warning that diffused open models may matter more than frontier capability reframes the AI race as a contest of architecture — and if he's right, the capital funneling into a handful of frontier labs is backing the wrong architecture.
Ask SkimNews
