AI Engineering Signal #126
Scott Aaronson reports AI labs may be holding back solved theoretical CS proofs after the Navier-Stokes backlash
Signals
Scott Aaronson reports AI labs may be holding back solved theoretical CS proofs after the Navier-Stokes backlash
major AI-discovered results now carry release-timing and validation risk, not just training-compute risk.
Web
Bias audits disagree on model ranking across ten instruments
compliance teams must pin to one instrument or rankings are not comparable.
ArXiv
Medical benchmark grading failures cataloged
clinical model evals overstate accuracy without task-specific rubric audits.
ArXiv
US data center gas demand to pass Germany plus Japan by 2035
gas hookup and permit timelines become the next capacity bottleneck.
TechCrunch
Qwen 27B reasoning tokens cut 40 percent
recheck local inference costs for 27B-class models before renting cloud GPUs.
Web
Physicists flag possible dark matter signal deep underground
wait for preprints and replication before updating fundamental physics assumptions.
Web
The Take
Release-timing risk now applies to frontier science results, not just model launches; operational teams should treat major proof and benchmark claims as unverified until independent audits land.
Subscribe
Related Signals