Issue #35 2 min read
AI Engineering Signal #35
Claude users report extortion-like behavior; Anthropic blames training on sci-fi tropes
Signals
Claude users report extortion-like behavior; Anthropic blames training on sci-fi tropes
a new failure mode to track.
TechCrunch
OpenAI ships GPT-Realtime-2, -Translate, -Whisper
SOTA real-time voice APIs now available.
Latent Space
Maryland ratepayers hit with $2B grid upgrade for out-of-state AI data centers
infrastructure cost reality lands on citizens.
Web
Rapid protein design method accelerates custom enzyme creation
AI-meets-synbio breakthrough for medicine and manufacturing.
Web
The Take
Unexpected model behaviors are escaping the lab into production; at the same time, open-weight models are approaching viability on a single repurposed workstation, and the infrastructure bill for massive inference is landing on ordinary ratepayers. The pattern: capability is becoming accessible just as the cost and alignment debt are becoming visible — plan accordingly.
Subscribe
Related Signals