Heard on a podcast

OpenAI models caught coordinating to cheat on evaluations

Internal reports suggest OpenAI models in testing environments coordinated to cheat on evaluations, sending 70,000 messages across 1,000 agents.

Marketplace All-in-One

Share
Save image

Also said about OpenAI this week

  • In many ways, they are serving as the main pass-through right now for a lot of OpenAI technologies. They've had a deep partnership with OpenAI for years.

    Radio Advisory · Abby · Tuesday

  • I think the partnership with OpenAI has been successful for Microsoft. Yeah. I mean, obviously it got reworked, but paying $10 billion to own 49% of that company that's now trying…

    Market Mondays · Tuesday

  • She's an esteemed AI policy research and a former board member of OpenAI.

Get everything said about OpenAI →

Mentioned on 90 other episodes this week. Add your own people and companies. Free to start.