I’ve joined Anthropic

Original Article ↗ Hacker News Discussion ↗

Motivations and Timing of the Move

  • Many speculate the key drivers are access to frontier-scale compute, talented teams, and being close to where “the real action” is, rather than pure cash.
  • Others think the IPO-era upside and “last great window” for researchers to get very rich in LLMs is a major factor.
  • Some argue if he only cared about money he’d start his own lab; others point out that running a lab means years of fundraising, hiring, and management he may not want.

Role and Technical Focus at Anthropic

  • Reports say he’s joining the pretraining team to lead work on using Claude itself to accelerate pretraining research (recursive, agentic test-time scaling in the spirit of his “autoresearch” projects).
  • Some are excited about pushing “LLMs optimizing LLMs”; others see current demos as glorified hyperparameter tuning, not qualitatively new research.

Impact on Anthropic and Perception

  • Widely seen as a big talent and branding win that reinforces Anthropic’s narrative as a frontier lab and stabilizing alternative to rivals, especially pre‑IPO.
  • There’s debate whether he’ll be primarily an R&D contributor or more of a high-prestige educator/influencer/DevRel figure that markets Claude by example.

Ethics, Safety, and Military Involvement

  • Heavy argument over Anthropic’s “good guys” branding:
    • One side cites safety red lines and earlier refusals to cross them as evidence of relative virtue.
    • Critics highlight policy rollbacks, quiet DoD work (e.g., Mythos, military targeting in Iran conflict), and see AI-safety rhetoric as PR and regulatory-capture strategy.
  • Broader moral disputes emerge over AI for war, “defending democracies,” and whether any large US/Chinese AI org can be considered ethical.

Evaluation of His Track Record

  • Many praise him as an exceptional educator and communicator who helped train a generation of ML practitioners.
  • Views on his technical and ethical record are mixed:
    • Some credit pioneering image–text work and key Tesla Autopilot techniques.
    • Others fault the “vision-only” self-driving bet and see moral complicity in deploying unsafe systems.
    • Recent “vibe coding”/agentic projects are seen by some as insightful, by others as overhyped or derivative.

Broader AI and Market Dynamics

  • Discussion broadens to Anthropic vs OpenAI vs Google vs Chinese open‑weight labs, fears of AI monopolies, regulatory capture, and job destruction in white‑collar, low‑code, and agency work.
  • Some believe open-source models plus cheap hardware will eventually erode closed‑lab moats; others point to explosive closed‑lab revenue as evidence that a moat still exists.
  • Several worry that Anthropic’s products (especially coding agents) are already accelerating white‑collar displacement and shifting power from labor to capital.

Education and Open vs Closed Tension

  • Many are disappointed his independent education startup appears paused and that he joined a closed LLM lab instead of backing open models.
  • Others hope proximity to frontier research will ultimately make his future educational content better, if NDAs and corporate priorities allow it.