@zephyr_z9 Does this mean the end of fixed subagents (like in Claude or Codex) and a shift to a main process that spins up agents on the fly to work in parallel, using RL rewards?
tldr SWIM: O(n) membership protocol. Each node probes one random peer per τ interval. Timeout → k indirect probes via random members. Still no ACK → suspect state → timeout → dead. Incarnation numbers resolve conflicting state claims. Dissemination piggybacks on PING/ACK/PING-REQ. No protocol phase separation.
https://t.co/NfeKx0EOSY
@devdotsasha@ky__zo Just found this actually https://t.co/x9cYPdAEj8. Amazing UI working both on desktop+phone and free version is integrated by git, looking good
@rauchg This is why the best AI users aren't necessarily the best engineers - they're people with deep domain expertise who now have infinite leverage. The technique is knowing what to build, not how to build it.
@zoink The lifespan heuristic is sharp. Throwaway scripts, one-off automations - let AI go wild. Production systems touching money or users? That's where the human judgment tax is worth paying.
Never disrespect the masculine urge of those who monitor the situation
Monitoring the situation is one of life’s great pleasures for those that love to hyper fixate on things
Ever wonder how billions of devices share the internet without it collapsing?
TCP congestion control. A 36-year-old algorithm running on every device, constantly asking: "how fast can I go?"
Built a TUI to visualize the famous "sawtooth" pattern:
#rust#Linux #SoftwareEngineering
If you care about latency, throughput, or tail performance, this matters. Most async runtimes and networking stacks are built on one of these as their backbone:
• epoll - readiness-based, kernel tells you when to act - you still do the syscall
• io_uring - submission/completion-based, you submit work up front - kernel returns completed results