OpenAI has reached an agreement with the Defense Department to deploy its models in the agency's network, company chief Sam Altman has revealed on X.

Sam Altman says “the DoW displayed a deep respect for safety.”

Not 24 hours ago, he seemed to back Anthropic “supporting our warfighters” as long as two “red lines” weren’t crossed, though his tepid support was laden with five instances of “I think” and one “mostly.”

The two “red lines” in question:

  • Domestic mass surveillance
    (presumably, foreign mass surveillance is ok)
  • Autonomous weapons
    (likely because they would be held legally liable for misfires)
    • XLE@piefed.socialOPEnglish
      9·
      10 hours ago

      From best to worst

      • Still preferably none, your brain is a muscle that needs exercise
      • A local model with beefy hardware, if you can install one. Otherwise, a less powerful model that’s still local, like Ensu.
      • A proxied model from a company with a good track record, like DuckDuckGo or Proton
      • Out of the ones you listed, DeepSeek might be the least bad. As long as you don’t send personal information to it, stick to the site and not a permission-hungry app, remain behind a VPN, and aren’t a recent expatiate from China asking prickly questions, it’s probably fine.
      • Google hates you. Gemma is Google. They are a part of the American surveillance apparatus incarnate.