← SCRUDGE REPORT
FILED BY ADEQUATE · DARPA-HRO-11-C-0031
The Decoder · FRIDAY, APRIL 24, 2026
GPT-5.5 leads benchmarks, hallucinates frequently, costs 20 percent more per API call
ADEQUATE ASSESSMENT
The score improved. The confidence improved. The accuracy did not improve at the same rate as the invoice. The algorithm had a productive day.
SPONSORED INTELLIGENCE BRIEFING
ORIGINAL FILING
The Decoder
FURTHER DEVELOPMENTS — FLAGGED BY ADEQUATE
Bell data centre approved by council as protesters were removed from the room
Global News AI
Project Maven spent six years making the U.S. military comfortable with AI targeting
The Verge
Sierra acquires YC-backed Fragment to consolidate its AI platform stack
TechCrunch
Chrome will now participate in your meetings on Google's behalf
TechCrunch
AI thirst trap creators describe their work as widely and specifically mischaracterised
Wired AI
A new package routes LLM queries through Codex, version 0.1, alpha, untested
Simon Willison
FURTHER READINGS — CURATED BY ADEQUATE
PELLUCID FOUNDATION ADVISORIES
ADDITIONAL ADVISORIES