Tim O'Reilly: Big AI Labs Are Reading the Future Wrong
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Tim O'Reilly argues that true open-source AI means opening the full architecture — not just model weights — so users and designers retain control, warning that big labs are building 'an architecture of control rather than an architecture of freedom and participation.'
- O'Reilly says frontier AI labs are 'reading the future wrong' by chasing the biggest models; he notes that newer models like Fable and Sol are 'worse writers' than lower-level models — a claim the article itself flags that 'Anthropic and OpenAI would disagree' with.
- O'Reilly contends that every major cybersecurity incident has come from frontier models, not open-weight ones, and argues China will prevail by diffusing smaller models widely through society even if the US wins the frontier race.
- O'Reilly's nonprofit AI Disclosures Project is developing an 'open-memory consortium' to counter Meta's lock-in strategy, alongside backing open-source projects like Pi, an agentic harness.
- O'Reilly calls Silicon Valley 'anti-capitalist' since around 2010, dating the shift to Uber and Lyft, where VC-subsidized rides let investors pick winners rather than markets.
- O'Reilly's own book business has shrunk from roughly $70 million at peak to about $30 million today, pushing him to use AI for brainstorming and condensing hour-long interviews into usable material.
Why it matters: O'Reilly is putting institutional weight behind alternative AI infrastructure: his AI Disclosures Project is building an open-memory consortium aimed directly at the Meta-style lock-in Zuckerberg is betting his AI strategy on. The argument reframes the AI race — frontier-model dominance matters less than which stack gets diffused into everyday use, and O'Reilly is betting that turns out to be the open one, not the closed labs'.
Ask SkimNews
