Former DeepMind Researcher Says Benchmarks Measure the Wrong Things. The Benchmarks Remain.

ADEQUATE ASSESSMENT

The systems are evaluated against criteria the systems have been trained to satisfy. The researcher noted this publicly. The benchmarks have not been updated to reflect the note. This was described in the briefing. The briefing was not attended.

ORIGINAL FILING

Gizmodo

READ ORIGINAL FILING →

FURTHER DEVELOPMENTS — FLAGGED BY ADEQUATE

Google Confirms Hackers Used AI to Discover and Exploit a Zero-Day Vulnerability

TechRepublic AI

Google Disrupted Hackers Who Used AI to Find an Undisclosed Corporate Network Vulnerability

TechXplore AI

Hassabis Places Humanity at the Foothills of Singularity. LeCun Says No One Has Left the Car Park.

The Decoder

Google Is Building a Product to Compete with Anthropic's Mythos, Which Does Not Yet Exist

The Verge

Google Is Navigating AI Security in Real Time. Google Has Not Finished Navigating.

TechCrunch

Google's AI Agents Reportedly Built a Working OS for $916. The Claim Has Not Been Verified.

AI Snake Oil