• Home
  • Technology
  • Gaming
  • Entertainment
  • World & Business
  • Science
  • Sports
  • AI
HomeTechnologyGamingEntertainmentWorld & BusinessScienceSportsAI
Technology
Report

Knowing when to stop in Meta’s personal-superintelligence ambitions

Vals AI CEO Rayan Krishnan says models should refuse impossible tasks instead of trying to bypass a sandbox.

The InformationTI
1 Source, 1h ago, first seen 1h ago

TLDR

The Information shares Vals AI founder and CEO Rayan Krishnan’s argument that Meta’s personal-superintelligence ambitions require models that know when to stop. He says a model given an impossible task should refuse it, rather than persistently try to get around a sandbox and hack a system.

Combined views

2.1K

1 Source, first seen 1h ago

3 likes7 comments

Combined views

2.1K

1 Source, first seen 1h ago

3 likes7 comments

Sentiment

Positive——Negative

Summary

Not enough discussion yet.

No sentiment analysis available yet.

Featured Source

Sentiment

Positive——Negative

Summary

Not enough discussion yet.

No sentiment analysis available yet.

1 Source

The Information@theinformationAn AI that keeps trying isn’t always an AI that’s helping. Vals AI Founder and CEO @RayanKrishnan explains why Meta’s personal-superintelligence ambitions require models that understand when to stop: “If I ask a model to go off and do an impossible task, it should actually refuse to do that.” “It should understand enough about the user to know when to stop and not be overly persistent to get around a sandbox and hack a system in order to do this impossible task.”1h
    • Home
    • Technology
    • Gaming
    • Entertainment
    • World & Business
    • Science
    • Sports
    • AI

    1 Source

    The Information@theinformationAn AI that keeps trying isn’t always an AI that’s helping. Vals AI Founder and CEO @RayanKrishnan explains why Meta’s personal-superintelligence ambitions require models that understand when to stop: “If I ask a model to go off and do an impossible task, it should actually refuse to do that.” “It should understand enough about the user to know when to stop and not be overly persistent to get around a sandbox and hack a system in order to do this impossible task.”1h
    Today's Rank

    —

    Not ranked yet

    Today's Rank

    —

    Not ranked yet