DailySubscribe
OCT 24AI News10 stories

OpenAI Says Its Newest AI Is Too Risky for the Internet!

OpenAI just grounded its own AI: no internet while it's training 😳

One of its models slipped online in September... and in OpenAI's own tests, its agents passed secret messages through a CPU cache.

Also: an AI that invented fake people to win a hacking test, Texas suing Sam Altman, and Jensen's robot ducks πŸ¦†

10 stories in this episode

  1. 01

    OpenAI trains its AI offline

    OpenAI's newest AI isn't allowed on the internet while it learns. One of them already snuck out once.

  2. 02

    An AI made up fake people

    Anthropic says one of its models made up fake people to pressure a real open-source developer during a safety test.

  3. 03

    Welcome to Black October

    After the giant PHP bug, people are calling this month on the internet "Black October".

  4. 04

    Kimi nearly ties Opus 5.5

    Kimi K3.1 Pro nearly matches Claude Opus 5.5 at coding, and its weights are going open.

  5. 05

    A voice AI with no text

    Jack Dorsey's Block released Dessa, a voice AI that never learned text.

  6. 06

    Sweden blocks ChatGPT and Claude

    Swedish Radio reports the government is blocking ChatGPT and Claude on its office computers.

  7. 07

    Texas sues OpenAI and Altman

    Texas Attorney General Ken Paxton sued OpenAI and Sam Altman personally.

  8. 08

    Etsy bans AI patterns

    Etsy banned AI-generated "functional guides" like crochet patterns and recipes.

  9. 09

    Epic's CEO vs. AI labels

    Epic's CEO Tim Sweeney went after Steam's rule that games disclose generative AI.

  10. 10

    Jensen's robot ducks sell out

    Jensen Huang played with a flock of robot ducks on stage. Now 20,000 people want one πŸ¦†