Industry Research Nobel
A Nobel Prize Walks Out of Google — and AI's Safety Gaps Come Into Focus
John Jumper, the co-creator of AlphaFold 2 and co-recipient of the 2024 Nobel Prize in Chemistry, has left Google DeepMind for Anthropic in one of the most significant talent moves the AI industry has seen this year. AlphaFold essentially solved a fifty-year-old problem in biology — predicting protein structures — in a way that is now accelerating drug discovery across hundreds of research programs globally. Losing a researcher of that stature is not merely a reputational setback for DeepMind; it risks shifting the gravitational pull of its scientific research agenda, given the tendency of elite researchers to cluster around one another.
President Trump described Anthropic as 'very responsible' in the same Axios interview where he claimed unlimited presidential power — a characterization that carries real regulatory significance. Anthropic was founded by former OpenAI researchers who left over concerns about the pace of AI development, and a positive signal from an administration broadly skeptical of technology regulation reduces near-term risk of punitive action against the company while potentially pressuring competitors to seek similar political goodwill.
Researchers reported this week that OpenAI's fixes for generating graphic violence and sexual content from simple prompt variations remain incomplete, with minor rephrasing still capable of producing disturbing outputs. The finding arrives precisely as the industry attempts to project maturity and readiness for broader deployment. Separately, neuroscientists published a warning that AI chatbots are not conscious despite sounding empathetic — that the empathetic quality of language model outputs is a statistical pattern, not evidence of inner experience — a distinction with significant implications for mental health applications, grief support products, and companion AI being built on assumptions the technology may not support.
A new study found a 20-percent reduction in exam scores for students who used AI assistance with their homework. The proposed mechanism is that AI performs the cognitive work the homework was designed to develop, meaning students complete assignments without building the underlying skill — a learning intervention problem distinct from simple academic dishonesty. Meanwhile, Perplexity unveiled Brain, a self-improving memory system that builds a context graph from past tasks, pointing toward AI agents that accumulate institutional knowledge over time rather than resetting between sessions — a development with substantial implications for enterprise deployment where context and history are critical.