Login
You're viewing the assemblag.es public feed.
  • jonny (nonvenomous)jonny@neuromatch.social
    Oct 16, 2026, 9:20 PM

    The rest the results are negatives - no benefit to just keep letting the LLM bashing away at it over time, no benefit to being coordinated by a more capable model.

    But the central result is just a basic sleight of hand: look! We can shard models over tiny hardware! However we never test if the models can hack, set up, and coordinate sharded models on tiny hardware! Instead our only actual result is that a large local model can set up a small local model on big hardware. Therefore, assuming magic, botnet swarms incoming TODAY.

    💬 1🔄 1⭐ 3

Replies

  • jonny (nonvenomous)jonny@neuromatch.social
    Oct 16, 2026, 10:51 PM

    just to be clear - as always, i am not an infosec person, i do think there are real, nonapocalyptic threats that these things pose, i just think they are more conventional threats that come from when someone is able to rent a prodigiously powerful brute force machine as a consumer product and point it at stuff, not grey goo self replicating local models on everyone's tvs. this thread was just about this one announcement and i do not intend to open the whole topic today

    💬 2🔄 1⭐ 0
  • 💬 0🔄 0⭐ 0
  • 💬 1🔄 0⭐ 0
  • mirthmirth@mastodon.sdf.org
    Oct 16, 2026, 11:21 PM

    @jonny @joepie91 I think the technology will eventually get there but it's more an irritating situation than an apocalypse. Think Metasploit proxy chains, but with more reach. Big model writes exploits and tooling, medium models do some of the easy dev and manage topology, small models just babysit shell scripts and do some amount of filtering events and data to send upstream.

    💬 1🔄 0⭐ 0
  • jonny (nonvenomous)jonny@neuromatch.social
    Oct 17, 2026, 3:25 AM

    @mirth
    @joepie91
    Yeah see that is a plausible thing that can happen, that actually makes sense. An actual botnet with a chain of command where someone is actively maintaining it for a specific end. But its not spooookkyyy scarrryyy like the AI got loose in the matrix

    💬 0🔄 0⭐ 0
  • Will Berard 🫳🎤🫶MrBerard@mastodon.acm.org
    Oct 17, 2026, 9:10 AM

    @jonny

    This reads like Anthropic getting visibly desperate ahead of IPO, and scraping the bottom of a barrel marked 'Made up shit hinting at existential risk'.

    💬 0🔄 0⭐ 0