when did the looney tunes drunk stork retire? do kids these days even know their origin story, or are parents just lying about having sex with each other
playing with agent societies. asked claude to visualize a religion experiment -- two agents fighting to convert a crowd of 2,880 souls. it drew this. we've achieved AGI
Hey @ClaudeDevs, can y’all make the Code section an accordion that displays threads or make “Recents” switch over to remote control sessions on selection?
vu retweeted
dataset looked fun so i took a stab at overlaying it with a few orthogonal datasets (demographics, elections, county geo)
pretty fun to see how much local law differs across county/state borders (two near-identical counties on opposite sides of a state line can look completely different)
civic-texture-web.vercel.app…
github: github.com/vu1n/civic-textur…
datasets stitched in:
- LOCUS (yours)
- ACS 5-yr demographics
- county presidential returns '08-'24
- census county geometry + a city→county crosswalk
telling claude to schedule your loops to run when your limits refresh overnight and waking up to a fucking permission prompt. i need to localmax
Meet Puck 🧚 — a digital squatter on your desktop, peeks at what you're doing, comments on it
a deliberately useless computer assistant, built for #BuildSmall @huggingface @Gradio
how many devs are out here rawdogging LLMs like this?
Interesting new SWE/agentic benchmark (DeepSWE) was released yesterday. 113 tasks across 91 repos in 5 languages. Here are interesting things I noticed:
- The evaluation harness (mini-swe-agent) gives every model a single bash tool and the same SI. No vendor editing primitives.
- Eval Prompts are shorter than SWE-Bench Pro, but require 5.5× more code and touch 7 files on average. The idea is to mimic how developers actually talk to agents, short behavioral descriptions, not verbose specs.
- SI describes a specific workflow: find code, reproduce, fix, verify, edge cases, submit. This maps directly onto how the verifier grades, which could bias toward models that follow instructions literally over models that explore more.
- The bash tool is guarded, outputs over 10k chars get truncated. Malformed tool calls get caught and retried with guidance rather than crashing. To prevent to blow up context.
- Mini-swe-agent claims to match or beat 1P harnesses on the same tasks. Claude Opus scored +10pp over Claude Code. Gemini 3.1 Pro scored +20pp over Gemini CLI.
Would love to see how other harness × model combinations will do, e.g. @cursor_ai, @antigravity, @FactoryAI and how well the eval harness does on more general knowledge work, e.g. GDPval.
Great to see the SWE-agent team keep pushing on both the research and eval side. 🤗
