AI Evaluation

A group of independent researchers working together around a table with laptops and notebooks in a bright office space.

OpenAI’s Safety Fellowship Is Not a Generic Grant: It Trades Model Access for Structured, Governance-Focused Research

admin2 months ago05 mins

OpenAI’s Safety Fellowship, announced on April 6, 2026, is best understood as a structured safety research program rather than a broad grant round. It combines funding, mentorship, compute, and optional Berkeley workspace for independent researchers, but it withholds internal system access and sharply targets work that could inform AI governance, safety standards, and enterprise risk…

LLMs Do Not Succeed in Games by Default. The Benchmark and API Layer Is Doing Much of the Work

admin2 months ago05 mins

Recent game-playing results from large language models are easy to overread. The stronger finding is not that LLMs can simply be dropped into games, but that their performance changes sharply when researchers add task-specific evaluation harnesses, game interfaces, and supporting modules that compensate for weak planning, action formatting, or memory. LMGAME-BENCH shows the gap between…

Google DeepMind’s AGI Framework Shifts the Debate From Bigger Models to Measured Cognitive Abilities

admin2 months ago05 mins

Google DeepMind is trying to make AGI progress harder to overstate. Its new framework replaces vague milestone talk and single benchmark scores with a structured test of ten cognitive abilities, then asks a stricter question: how those abilities combine, and how the result compares with demographically representative human baselines. Ten abilities instead of one headline…

Codex Is Not Replacing Finance Reporting Systems; It Is Taking Over the Manual Drafting and QA Around Them

If Assistive Robots Are Going to Leave the Lab, Stretch 4 Shows What Has to Change First

ChatGPT at 900 Million Weekly Users Signals Two Markets Moving at Once

AI Inference Chips and AI-Native Wi-Fi Are Advancing Together, Not Separately

If a Campus Can Enforce AI Rules and Keep the Network Stable, OpenAI’s Student Club Push Becomes More Than Outreach

Orbital AI Data Centers in Space Are Now a Real Test Case, Not a Near-Term Replacement for Earth

Robot Hand Dexterity Is Moving on a Different Curve Than Generalist AI

As Codex Moves From Code Suggestions to Code Execution, OpenAI’s Security Model Gets Much More Granular

OpenAI’s GPT-5.5-Cyber rollout starts with access tiers, not a jump in autonomous hacking

Why Sardinia’s coal exit still hinges on trust, not just wind, solar, and cables

OpenAI’s Safety Fellowship Is Not a Generic Grant: It Trades Model Access for Structured, Governance-Focused Research

LLMs Do Not Succeed in Games by Default. The Benchmark and API Layer Is Doing Much of the Work

Google DeepMind’s AGI Framework Shifts the Debate From Bigger Models to Measured Cognitive Abilities