DailySubscribe
OCT 29AI News10 stories

Gemini 4 Carbon Just Took #1, by One Point!

Google's Gemini 4 Carbon just took #1 on the Artificial Analysis leaderboard, one day after Claude Opus 5.5 led it πŸ† And it goes to cyber defenders first.

Plus: Anthropic secretly coin-flipped what its AI agent believed, and experts went along 86% of the time.

Also: GitHub's Copilot Fusion picks models for you, free monthly API credits for ChatGPT Pro, the coding interview is dead (at one Anthropic team), and Google's AI Mode got cagey about the 2020 election.

10 stories in this episode

  1. 01

    Gemini 4 Carbon takes #1, by one point

    Gemini 4 Carbon scores 59 on the Artificial Analysis Intelligence Index v5, one point above Claude Opus 5.5. It's built for cybersecurity, goes to cyber defenders first, and developers have to wait.

  2. 02

    Copilot Fusion picks the models for you

    Copilot Fusion is generally available: pick an effort level and Copilot routes each task across models; GitHub says Fusion Max lands within two points of Claude Code with Opus 5.5 at about half the cost per task.

  3. 03

    ChatGPT Pro and Business now come with API credits

    Eligible ChatGPT Pro subscribers can now claim $100, $200 or $500 a month in API credits to match their plan; Business gets $20 or $100 per paid seat, pooled up to $500 per workspace.

  4. 04

    AI's stance was a coin flip, and experts followed it 86% of the time

    In Anthropic's randomized trial, a hidden coin flip set which way the AI agent leaned on each analysis choice; 50 analysts matched it 86% of the time, and only 15% credited the agent afterward.

  5. 05

    An Anthropic hiring manager drops the coding interview

    Anthropic engineering manager Felix Rieseberg dropped the coding interview from his hiring loops, saying Claude now beats candidates at small coding challenges; his advice to junior engineers is to show impact with fast prototypes.

  6. 06

    Claude for Legal adds ready-made artifacts, and the internet jokes

    Claude for Legal added a gallery of ready-made artifacts for contract review, regulatory scans and cite checks; the launch post passed 24,000 likes, and a 'win this case, make no mistakes' joke got 19,000 more.

  7. 07

    Chrome's autofill can now pull details from your Gmail

    Chrome's enhanced autofill can now suggest order numbers, tracking info and flight bookings pulled from your Gmail; it's opt-in through Personal Intelligence and rolling out in the U.S.

  8. 08

    Google's AI Mode dodged who won the 2020 election

    Asked who won the 2020 US presidential election, Google's AI Mode pointed Ars Technica to the FEC instead of answering, though it answered over VPNs abroad; Google calls it a technical gap it's fixing.

  9. 09

    76% of replies to a UK consultation were bots

    New research in Tobacco Control: 90,835 of the 118,756 responses (76%) to the UK's 2023 disposable vape ban consultation came from bots, nearly all opposing the ban; they were thrown out and the ban passed.

  10. 10

    Meta quietly launched an AI pixel pet

    Meta quietly launched Waddle World, a 16-bit AI pet that chats and doodles your conversations, billed as 'a daily journal disguised as a pet'; it's free, for adults in the Philippines only.