Tag

agent-architecture

1 issue found

Nov 2, 2026

Description

Reality Check SemiAnalysis's GDPx puts the best model at just 12% on real chip design, cyber, and bio work; Gaia2/ARE reports frontier agents near 42%.

Security Gap Poisoning 0.1% of pretraining data reportedly plants backdoors that survive post-training; an agent intrusion postmortem counts ~17,600 actions.

Build Signal DeepSeek open-sourced a super-agent harness — verified actions, not retrieval, remain the bottleneck.

Goal Cup Starts 7 Nov One goal, one prompt, and your agent builds alone for up to 168 hours. Sign up free →