AI Engineering Signal #64
John Jumper, AlphaFold architect and Nobel laureate, leaves DeepMind for Anthropic
Signals
John Jumper, AlphaFold architect and Nobel laureate, leaves DeepMind for Anthropic
signals Anthropic is building serious scientific research capacity beyond LLM scaling.
TechCrunch
OpenAI Codex logging bug writes terabytes to local SSDs
audit disk usage on any machine running Codex agents before storage fills.
GitHub
GLM-5.2 benchmarks competitive with frontier models
evaluate as an open-weight routing alternative before renewing Claude Sonnet 4.6 or GPT-5 contracts.
Web
Anthropic identity verification confirmed as age-gating only
no API access change, but biometric ID is now in the consumer auth path.
Web
Sakana AI releases Fugu open foundation model
adds a Japan-origin open-weight option where export-control exposure affects procurement.
Web
AI-generated Amazon e-books tripled since late 2022
RAG corpora or training sets sourced from Amazon need provenance filtering now.
The Take
Frontier capability is no longer a two-lab moat, and the researchers who built the most consequential models are moving. Model selection is now a quarterly routing audit, not a one-time procurement call.
Subscribe
Related Signals