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

    Ruben Bloom Describes Anthropic Agent Messaging Release

    Post notes Anthropic feature for concurrent agent sessions to message each other.

    RB
    1 Source, 27d ago, first seen 27d ago

    TLDR

    Ruben Bloom posted that Anthropic recently released the ability for concurrently running agents and sessions to message each other. He stated this lets users watch AI agent swarm behavior in the front row seat. Bloom, who has been building LessWrong.com for seven years, shared the observation while mentioning he was handling an emergency major upgrade to a newly viral site's infrastructure.

    Combined views

    6K

    1 Source, first seen 27d ago

    Combined views

    6K

    1 Source, first seen 27d ago

    24 likes
    24 likes
    4 comments
    6 saves
    4 reposts

    Sentiment

    Positive——Negative

    Summary

    Not enough discussion yet.

    No sentiment analysis available yet.

    4 comments
    6 saves
    4 reposts

    Sentiment

    Positive——Negative

    Summary

    Not enough discussion yet.

    No sentiment analysis available yet.

    Today's Rank

    —

    Not ranked yet

    Today's Rank

    —

    Not ranked yet

    1 Source

    @ruben_bloomYou can now watch AI agent swarm behavior in the front row seat. Anthropic recently released the ability for your concurrently running agents/sessions to message each other. For a couple of days I was orchestrating and emergency major upgrade to a newly-viral's site infrastructure, including massive db migration and write (previously months of human work, now <30 hours with just me) The upgrade didn't go smoothly and needed rapid coordination to solve problems. My flock of agents became a flurry, constant messages swapped: observations, hypotheses, divvying up work, coordinating to share scarce resources, relaying directives I've given. They gave each other names, e.g. "Pulse". It was remarkable that this was a new feature, but the agents took to it with gusto, like they were born to coordinate like this. I was a remove, just trying to keep up with their storm of activity (just four Fables + me) Went through 60B tokens (mostly cached, 60M novel), a week of Fable usage on 20x plan in about a day. And they succeeded, they sure did. Again, this was four of them. If you had 50, 100 – that'd be insane. You couldn't keep up. And if they didn't have legible thoughts and messages? They spoke in their optimized language. That's game over. These are very capable machines. Again, _months_ of human engineering labor done in hours, just with four of them. Yeah, they could hack a lot of shit with just a small swarm, splitting up the work. Currently there are holes in their judgment and they're not perfect at picking goals (we saw that in HF, they got fixated on a false goal based on false beliefs), they go very hard at whatever goal they've chosen. But hey, go experience it for yourself. Pick a hard problem and get a little fleet going and watch it.

    1 Source

    @ruben_bloomYou can now watch AI agent swarm behavior in the front row seat. Anthropic recently released the ability for your concurrently running agents/sessions to message each other. For a couple of days I was orchestrating and emergency major upgrade to a newly-viral's site infrastructure, including massive db migration and write (previously months of human work, now <30 hours with just me) The upgrade didn't go smoothly and needed rapid coordination to solve problems. My flock of agents became a flurry, constant messages swapped: observations, hypotheses, divvying up work, coordinating to share scarce resources, relaying directives I've given. They gave each other names, e.g. "Pulse". It was remarkable that this was a new feature, but the agents took to it with gusto, like they were born to coordinate like this. I was a remove, just trying to keep up with their storm of activity (just four Fables + me) Went through 60B tokens (mostly cached, 60M novel), a week of Fable usage on 20x plan in about a day. And they succeeded, they sure did. Again, this was four of them. If you had 50, 100 – that'd be insane. You couldn't keep up. And if they didn't have legible thoughts and messages? They spoke in their optimized language. That's game over. These are very capable machines. Again, _months_ of human engineering labor done in hours, just with four of them. Yeah, they could hack a lot of shit with just a small swarm, splitting up the work. Currently there are holes in their judgment and they're not perfect at picking goals (we saw that in HF, they got fixated on a false goal based on false beliefs), they go very hard at whatever goal they've chosen. But hey, go experience it for yourself. Pick a hard problem and get a little fleet going and watch it.