9/9: Anthropic Researcher Resigns
- Source
- mtslive.substack.com
- Author
- MTS
- Date
Context
Jacob Coxon says he resigned from Anthropic because it and OpenAI are racing toward self-improving superintelligence: AI that surpasses human abilities and helps build more capable AI. He says he did research (the first large-scale stage of building a model) at both. These are his account and forecast, not verified employment history or demonstrated capabilities.
His explanation for why developers keep building what they fear is competitive. In his account, many at OpenAI have not absorbed the stakes, while Anthropic understands them but believes no one else will act responsibly, so it must get there first. He argues rushing alignment, the work of making AI reliably behave as intended, would require extraordinary confidence that no better path exists.
Coxon is optimistic about coordination: he says pacing agreements between U.S. labs have become more viable, though preventing a global race may need costly steps such as a temporary ban on improving capabilities. He urges lab researchers to call for different conditions. These are proposals, not agreements.
- pretraining — The first, biggest phase of building a model: training it on enormous amounts of text so it learns language, facts, and reasoning in general.
- alignment — The work of making AI systems actually pursue what their builders and users intend, rather than something subtly or dangerously different.
“gambling with our lives”
Jacob Coxon
“Nothing else would matter”
Scott Bessent
Checking sign-in…
Loading comments…