Live Feed
Seven Days After OpenAI's Clinical Benchmark Dropped, a Healthcare Specialist Topped It: Corti 60.5, ChatGPT for Clinicians 59.0
Corti Symphony scored 60.5 on HealthBench Professional, OpenAI's own physician-authored benchmark — surpassing ChatGPT for Clinicians (59.0) and GPT-5.4 with extended reasoning (48.1). On the adversarial red-teaming slice, Corti scored 87.7. GPT-5.4 scored 30.3. Physicians scored 30.
Microsoft's 100-Agent AI Harness Found 16 Windows CVEs — Including an Unauthenticated IKEv2 RCE
MDASH, Microsoft's multi-model agentic scanning harness, found 16 vulnerabilities patched in May Patch Tuesday — including 4 critical RCE flaws. It topped the CyberGym public benchmark at 88.45%, five points above the next entrant, and achieved 100% recall on confirmed Windows kernel bugs.
Recursive Superintelligence Raises $650M at $4.65B — GV, Nvidia, and AMD Back a Lab With 25 Staff and No Product
Richard Socher and Tim Rocktäschel's self-improving AI startup emerged from stealth with $650M led by GV and Greycroft, with Nvidia and AMD participating. Founded in late 2025, Recursive has not shipped a public product. The bet: that automating the AI research loop itself is the fastest path to superintelligence.
AI Data Centers Are Competing With Residents for Power — and Utilities Are Choosing Data Centers
Sierra Pacific Power is weighing a transmission line redirect that would serve a new AI data center at the expense of 50,000 Lake Tahoe residents. The same conflict is unfolding in Montana, San Antonio, and Washington State, with residential ratepayers consistently on the losing side.
Arena Activates Battles-in-Direct Votes — Two-Month Backfill Could Shuffle Top-10 ELO
Arena's experimental Battles-in-Direct program began counting toward official ELO rankings on May 12. A 30-day backfill of March through May votes is rolling out now, and the distributional shift toward longer, harder prompts may reshuffle models currently separated by noise-level margins.
Sutskever's $7B OpenAI Stake and 52-Page Altman Dossier Surface in Musk Trial
Ilya Sutskever disclosed an OpenAI stake worth $7 billion on the stand, then testified he spent a year building a 52-page evidence file on Sam Altman's 'consistent pattern of lying.' Closing arguments arrive Thursday.
Google Kills the Cursor, Launches Googlebook — Chromebook's Successor Ships This Fall With Gemini Baked In
At The Android Show on May 12, Google announced Googlebook, a new laptop platform built on Android and ChromeOS, with a Gemini-powered Magic Pointer replacing the traditional cursor. Partners include Acer, ASUS, Dell, HP, and Lenovo. Devices ship fall 2026.
Judgment Labs Raises $32M for Agent Evaluation Infrastructure — Lightspeed Returns in Under Six Months
Judgment Labs closed combined seed and Series A funding to build continuous improvement infrastructure for AI agents. Lightspeed Venture Partners led both rounds, returning less than six months after the initial seed — an unusually fast double-down on a company whose product is already in production at agent-native startups.
Sora 2 Pro Arrives on OpenRouter at $0.30/Second — OpenAI's API Video Strategy Takes Shape
After shutting down the Sora consumer product on April 26, OpenAI's Sora 2 Pro is now accessible via OpenRouter at $0.30 per second. At 1080p with synchronized audio and C2PA provenance, it targets production pipelines at six times the price of Google's Veo 3.1 Lite.
AI Venture Funding Hits $255.5B in Q1 2026 — More Than All of 2025 in a Single Quarter
PitchBook's Q1 2026 data shows the AI sector deployed $255.5 billion in three months, eclipsing the full-year 2025 total of $254.4 billion. Horizontal platforms took $197 billion across 396 deals. Autonomous machines hit $29 billion, driven by Waymo's $16B Series D.
Arena Adds Battles-in-Direct to ELO — Position Bias and Org-Context Advantage Now Corrected
Arena's May 12 update feeds direct-chat battles into leaderboard scoring for the first time, tightens confidence intervals, and explicitly corrects two newly discovered biases: Model A position preference and same-org context advantage.
Isomorphic Labs Closes $2.1B Series B to Scale AI Drug Design Engine
Demis Hassabis's AI drug design company closes a $2.1B Series B led by Thrive Capital, with Alphabet, Temasek, UK Sovereign AI Fund, and MGX participating. Capital goes toward IsoDDE deployment and a clinical-stage drug pipeline.
Perceptron Mk1 Launches at $0.15/M: VSI-Bench 88.5, 80-90% Cheaper Than Frontier Video AI
Perceptron AI ships Mk1, a video understanding and embodied reasoning model that posts the highest VSI-Bench score on record and prices at $0.15/$1.50 per million tokens, against GPT-5's $2.00 blended and Gemini 3.1 Pro's $3.00.
Amp Raises $1.3B to Build an Independent AI Grid — Pooled Compute for Frontier Labs That Can't Afford to Own It
Amp PBC, a public benefit corporation founded by former a16z Oxygen chief Anjney Midha, has raised $1.3 billion to create a shared compute grid for independent frontier AI teams. The thesis: the most productive AI labs waste 30-40% of their FLOPs because they can't smooth their own demand curve.
Amazon Employees Are 'Tokenmaxxing' MeshClaw to Game Internal AI Leaderboards
Amazon rolled out MeshClaw — an internal agent platform inspired by OpenClaw — and tied developer AI usage to visible leaderboards. Employees responded by using the tool to automate useless tasks and inflate token consumption. Meta employees are doing the same thing.