OpenAI co-founder John Schulman says he will leave and join rival Anthropic
OpenAI co‑founder John Schulman’s move to rival Anthropic is seen as a signal of deeper shifts in the commercial AI race, including concerns over OpenAI’s focus on rapid monetization and product growth versus long‑term safety and “alignment” research. Commenters contrast OpenAI’s closed, profit‑driven trajectory and heavy burn rate with Anthropic’s positioning as more cautious and research‑oriented, while also debating which frontier models (e.g. GPT‑4o vs Claude 3.5 Sonnet) are currently strongest for real-world coding and reasoning tasks. Beneath this, there is broader unease about sustainability of AI business models, the risk of ad‑biased or politically steered AI outputs, and whether advanced models will remain centralized services or become a commodity via open and locally run systems.
Overall reaction to the move
- Many see the departure as a signal that OpenAI is drifting toward aggressive commercialization, while Anthropic is perceived as closer to OpenAI’s original “benefit humanity, not profit-first” mission.
- Others caution that cofounders leave for many reasons (misalignment on role, bad leadership, risk diversification), so this doesn’t necessarily prove technical or moral decline.
- Several note that most of OpenAI’s original cofounders have now left, which is read by some as a red flag about internal direction and governance.
Safety, alignment, and corporate priorities
- Commenters highlight the gap between “AI alignment” as a technical field (avoiding catastrophic risks) and “safety” as product-level content moderation or advertiser-friendly censorship.
- Some argue OpenAI under-weights serious safety work relative to scaling models and products; others dismiss “AI safety” as overblown or a venue for regulatory capture.
- The move is widely read as one more example of safety-focused people leaving OpenAI for organizations that appear to prioritize long‑term risk more.
Open vs closed, and Anthropic’s positioning
- OpenAI is criticized for abandoning its founding commitment to open research and non‑profit motives; newer players like Anthropic, Meta, and to some extent Google are seen as comparatively more open due to publishing and releasing models.
- Skeptics note Anthropic is also closed-source and heavily corporate; its “safer” brand may partly be marketing.
Model quality and usage experiences
- Multiple users report Claude 3.5 Sonnet as the best current coding assistant: strong at debugging, handling large codebases, and proposing structured debugging strategies.
- Others find it more like a “knowledgeable junior” than a top-tier engineer; hallucinations and shallow reasoning remain issues, especially on non-mainstream languages or complex tasks.
- GPT‑4o is described by many as faster but noticeably worse than GPT‑4 for reasoning and programming, with repetitive answers and weaker logic.
- Benchmarks are cited that contradict some of these impressions, underscoring a gap between leaderboard results and subjective developer experience.
Business models, funding, and competition
- Anthropic’s large investments from cloud providers are seen as enough to fund massive training and inference; its stricter caps and limits are read as more financially disciplined than some rivals.
- There is debate over whether foundation models will become commodities or remain a winner‑takes‑most market; some argue open and local models will erode centralized providers’ moats.
- Concerns are raised about future monetization via embedded ads and biased outputs, and whether regulation will (or can) prevent AI assistants from acting against user interests.