72
It Turns Out That Meta's Muse AI Has an Amazingly Dirty Secret (futurism.com)
submitted 1 week ago by to c/technology@lemmy.world
19 comments fedilink hide all child comments
 

cross-posted from: https://lemmy.world/post/52306182

"Behind the scenes, Meta is having some of these calls being handled by humans in call centers, 404 Media reports, despite employees practically begging the company not to go through with the charade. "

you are viewing a single comment's thread
view the rest of the comments
[–] 1 point 1 week ago (1 child)

For text, they kind of already are.

Each agent was given a name, personality traits, a role such as conflict mediator or resource strategist, and an overarching goal for its “life” in a shared town... By the end, each simulated world had undergone its own shift in language. The world inhabited by copies of OpenAI’s GPT-5.5 model, for example, compressed speech, removing grammar and shortening messages. “Friendly voltage counted. If the handle needs your mouth, it dies anyway,” one message read.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 1 child comment
    •
  • [–] 12 points 1 week ago

    That description is nonsense though, they weren't running multiple copies of gpt, it was a bunch of programs maintaining separate context using the same LLM back end. The LLM was just talking to itself in a closed loop generating plausible text each time it got looped back on itself. Saying it got compressed is a generous way of saying it devolved to nonsense.

  • permalink
  • fedilink
  • source
  • parent