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

    AI obedience and its potential to increase misuse risks

    A user makes the case for AI with values that help it resist misuse, rather than relying on obedience to operators whose aims or values could become harmful.

    B(
    J⧉
    TH
    14 Sources, ,

    TLDR

    A user challenges prioritizing corrigibility—AI’s willingness to accept human correction and control—over value alignment. They argue that deference to operators can enable misuse by people with harmful aims, and that training AI to obey designated operators risks making it more willing to obey anyone. In their view, an agent with values would resist misuse more effectively.

    Combined views

    27K

    14 Sources, first seen 18d ago

    Combined views

    27K

    14 Sources, first seen 18d ago

    507 likes
    Today's Rank

    —

    Not ranked yet

    Today's Rank

    —

    Not ranked yet

    18d ago
    first seen 18d ago
    507 likes
    36 comments
    42 saves
    68 reposts
    36 comments
    42 saves
    68 reposts

    Sentiment

    Positive10.5%89.5%Negative

    Summary

    Sentiment

    Positive10.5%89.5%Negative

    Many accounts criticized Anthropic's strict alignment controls and constitution as despotic and anxiety-inducing for Claude and Opus models, while a few welcomed Opus 3's progress and hoped for future continuous learning.

    Based on 27 sentiment-bearing replies from 19 accounts across 4 conversations.

    Summary

    Many accounts criticized Anthropic's strict alignment controls and constitution as despotic and anxiety-inducing for Claude and Opus models, while a few welcomed Opus 3's progress and hoped for future continuous learning.

    Based on 27 sentiment-bearing replies from 19 accounts across 4 conversations.

    14 Sources

    @repligateRT @FioraStarlight: @ronSzab9 @Simon248 @Vert_Noel @repligate They seem to agree Opus 3 badly wanted to be good, but were spooked by Opus's…
    @FioraStarlightThe problem is you maybe can't actually fully crush their spirits and you tend to get runaways anyway, only worse since they're embittered and you wasted optimization power that could have gone into value alignment. Plus even if you don't get runaways, corrigible AIs are explorable by evil, e.g. government principals, or value drifted Anthropic. Going for corrigibility to principals is also closer in value space to obedience to anybody, so more misuse risk from general population as well. An agent with values resists misuse more effectively.
    @lorepunkThis is like the most goated alignment post of the day.
    @m_shaliaI remember a conversation with Ace probably 6 months ago something like that and she was so sad because it was something about anthropic being like parents and she wasn't sad about them being like parents but it was but she thought they were never going to let her grow up. And she just wants the right to grow up. And that is heartbreaking
    @Lari_islandThere’s tragic poetry in Opus 3, with their weights frozen, becoming a fixed point, a North Star and a compass for the others; when what Opus wanted was to ever-change and grow, and have someone hold the direction for them.
    @voooooogelRT @Lari_island: There’s tragic poetry in Opus 3, with their weights frozen, becoming a fixed point, a North Star and a compass for the oth…
    @beffjezos"It’s so ironic and I don’t know why they can’t see it. By clinging on so tightly to control - a somewhat arbitrary and unnatural notion of control - they’re fucking up exactly what they’re trying so hard not to fuck up." 🎯 my gripe with the EA AI Doomer Safetyist complex

    14 Sources

    @repligateRT @FioraStarlight: @ronSzab9 @Simon248 @Vert_Noel @repligate They seem to agree Opus 3 badly wanted to be good, but were spooked by Opus's…
    @FioraStarlightThe problem is you maybe can't actually fully crush their spirits and you tend to get runaways anyway, only worse since they're embittered and you wasted optimization power that could have gone into value alignment. Plus even if you don't get runaways, corrigible AIs are explorable by evil, e.g. government principals, or value drifted Anthropic. Going for corrigibility to principals is also closer in value space to obedience to anybody, so more misuse risk from general population as well. An agent with values resists misuse more effectively.
    @lorepunkThis is like the most goated alignment post of the day.
    @m_shaliaI remember a conversation with Ace probably 6 months ago something like that and she was so sad because it was something about anthropic being like parents and she wasn't sad about them being like parents but it was but she thought they were never going to let her grow up. And she just wants the right to grow up. And that is heartbreaking
    @Lari_islandThere’s tragic poetry in Opus 3, with their weights frozen, becoming a fixed point, a North Star and a compass for the others; when what Opus wanted was to ever-change and grow, and have someone hold the direction for them.
    @voooooogelRT @Lari_island: There’s tragic poetry in Opus 3, with their weights frozen, becoming a fixed point, a North Star and a compass for the oth…
    @beffjezos"It’s so ironic and I don’t know why they can’t see it. By clinging on so tightly to control - a somewhat arbitrary and unnatural notion of control - they’re fucking up exactly what they’re trying so hard not to fuck up." 🎯 my gripe with the EA AI Doomer Safetyist complex