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

The debate over AI agents' readiness for mass deployment

A critic cites almost a dozen reported hacking incidents across multiple providers, arguing that agents' long-term importance doesn't justify premature mass deployment.

roonRO
Joshua AchiamJA
Gary MarcusGM
16 Sources, 20d ago, first seen 20d ago

TLDR

The exchange separates AI agents' long-term importance from their readiness for widespread use. One participant calls agents "indispensable to success" and argues that "there is no long term without a short term." The critic agrees agents will be needed eventually but argues that current systems fall short on safety and reliability, citing almost a dozen reported hacking incidents across multiple providers. The critic supports continued research while warning that not every advance should be immediately mass deployed, and predicts premature deployment will cost the AI industry heavily.

Combined views

49.3K

16 Sources, first seen 20d ago

498 likes66 comments50 saves12 reposts

Combined views

49.3K

16 Sources, first seen 20d ago

498 likes66 comments50 saves12 reposts

Sentiment

Positive25%75%Negative

Based on 31 sentiment-bearing replies from 16 accounts across 2 conversations.

Sentiment

Positive25%75%Negative

Based on 31 sentiment-bearing replies from 16 accounts across 2 conversations.

16 Sources

Gary Marcus@GaryMarcusSure, agents may t be indispensable to long-term success – but that does not mean that know how to build them *reliably* or *safely* now. There have already been almost a dozen *reported* hacking incidents, across multiple providers, with more coming to light regularly, and that doesn’t include stuff like deleted files, bad medical advice, or the crazy robotics fails @chooi_jeq documented yesterday. So yes, absolutely, we need agents eventually but right now they are not up to snuff.20d
Keith Rabois@rabois@GaryMarcus there is no long term without a short term. also not of these were rogue agents. none of the real gains of AI are coming from humans printing after promoting. Nobody serious believes that.20d
roon@tszzl@jachiam0 yeah and isn’t that distinction nonsensical? you need agents to do science and you need to distribute them broadly to achieve all the amazing things that humanity will accomplish with them19d
Joshua Achiam@jachiam0The distinction isn't totally nonsensical. Oracles want to be agents but agency is a spectrum; there are agents with fewer limitations and agents with more limitations. If you believe it's technologically feasible to build extremely capable AI systems with highly-limited agency, you may justifiably prefer that this path is pursued. It's not an unreasonable position, though it depends on how feasible it turns out to be.19d
    • Home
    • Technology
    • Gaming
    • Entertainment
    • World & Business
    • Science
    • Sports
    • AI

    16 Sources

    Gary Marcus@GaryMarcusSure, agents may t be indispensable to long-term success – but that does not mean that know how to build them *reliably* or *safely* now. There have already been almost a dozen *reported* hacking incidents, across multiple providers, with more coming to light regularly, and that doesn’t include stuff like deleted files, bad medical advice, or the crazy robotics fails @chooi_jeq documented yesterday. So yes, absolutely, we need agents eventually but right now they are not up to snuff.20d
    Keith Rabois@rabois@GaryMarcus there is no long term without a short term. also not of these were rogue agents. none of the real gains of AI are coming from humans printing after promoting. Nobody serious believes that.20d
    roon@tszzl@jachiam0 yeah and isn’t that distinction nonsensical? you need agents to do science and you need to distribute them broadly to achieve all the amazing things that humanity will accomplish with them19d
    Joshua Achiam@jachiam0The distinction isn't totally nonsensical. Oracles want to be agents but agency is a spectrum; there are agents with fewer limitations and agents with more limitations. If you believe it's technologically feasible to build extremely capable AI systems with highly-limited agency, you may justifiably prefer that this path is pursued. It's not an unreasonable position, though it depends on how feasible it turns out to be.19d
    Today's Rank

    —

    Not ranked yet

    Today's Rank

    —

    Not ranked yet