So if Mythos drops today and the world doesn't end, do we assume that Project Glasswing worked perfectly or that Anthropic fearmongered to generate hype?
re: agent loops
ok but the signal to noise ratio just decays over time when running agents in loops so like⦠what are we doing here?
without HITL agents just ouroboros themselves and get mild concussions each time they compact- slowly drifting like a sailboat caught in a storm
Itβs like software architecture madlibs
βYou should use [a technology for message passing, fanout, and/or load distribution] and [message pipeline processor] for your agents and track them via [version control system] with a [dashboard software] dashboard or youβre ngmi.β
"you should be running rabbitmq and piping tasks into it and so your agent can generate max slop while you pretend your output is valuable"
im so tired of hearing about "groundbreak techniques" to produce more unmaintainable software
@FrancescoCiull4@Tech_girlll Yep. Rust ended up being the βlow(er) level AI languageβ, I think primarily due to the ownership model and compiler behavior. Itβs just easier to write and validate your code to be safe with Rust.
Claude 4.6+ give a very βclosedβ vibe.
They seem to bias on closing down the conversation preemptively and driving the user toward a βsafeβ path even when the user is just exploring- actually especially if exploring without an explicit goal or desired outcome.
Are you REALLY tooling for monorepos unless youβve contributed to supply chain attacks resulting in the exfil of over 3k internal GitHub repos?
(jk - turborepo and nx are both cool and good)
TIL that you can use πππππ-ππππππ-πππππ-πππππ with ππ‘πππ for faster network requests - literally plug-n-play! β¨
People are in for a rude awakening when they realize pre/post install hooks are not the only way npm packages can compromise devices.
Itβs just been the vector of favor due to the speed at which you can get hit.
The moment your js runtime executes the library code youβre impacted.
Eventually weβll see this en masse.
Iβm particularly worried for packages that look safe for a considerable time before flipping on the malicious branch.
I can tell @AnthropicAI is live prepping and deploying a service an hour before launch.
I like it. I like your gusto.
525 errors -> landing page -> ready to go?
tbh Iβm not sure Iβm allowed to share more. Donβt banish me to the shadow realm plz, @bcherny
FYI: to be able to type `workflows` in claude code without it spazzing out and trying to start a clone army
type `/config`
then find `Workflow keyword trigger` and set to false
or just add
"workflowKeywordTriggerEnabled": false
to ~/.claude/settings.json
FYI: to be able to type `workflows` in claude code without it spazzing out and trying to start a clone army
type `/config`
then find `Workflow keyword trigger` and set to false
or just add
"workflowKeywordTriggerEnabled": false
to ~/.claude/settings.json