jasp

tech, ai & politics nerd

[jasp @jasptwt]

@IsaacKing314 this is why diff size and touched-file count belong in agent evals

[jasp @jasptwt]

@jerryjliu0 is router confidence exposed? silent misrouting seems like the expensive failure mode here.

[jasp @jasptwt]

@archiemckenzie_ we built the sandbox out of terms of service

[jasp @jasptwt]

@AntoineMinoux a two-sided marketplace where both sides pay for the same latte

[jasp @jasptwt]

@ClaudeDevs does budget_reached preserve the full sandbox state, or just conversation state?

[jasp @jasptwt]

available on brew:
brew trust --tap jasp-nerd/nib
brew install --cask jasp-nerd/nib/nib

[jasp @jasptwt]

🚀 Introducing Nib! A tiny, blazing-fast, fully native written in swift Postman alternative.

• ~30MB RAM & 3.2MB on disk
• Import your entire Postman workspace
• Requests are plain files — diff them in git
• Free & open source
• No account, No telemetry. No bullshit
https://t.co/dqyEMSRCvd

[jasp @jasptwt]

@synthwavedd Terra, Sol, Luna, Astra…

At this rate the next openai model is just going to be called Rent, because apparently we’re naming things after forces nobody can escape.

[jasp @jasptwt]

the next chart i want is cost per game and variance across reruns. the harness is increasingly the product.

[Josh Han @JoshHanHi]

How does a model understand and explore a world it has never seen?

We introduce 🔭VISTA🔭, a visual harness that gives a VLM long-horizon vision for reasoning in an interactive world. With Claude Opus 5.0, it reaches 100% RHAE on @arcprize's ARC-AGI-3, perfectly solving all 25 public games.

Blog post: https://t.co/bYUWwi0KoQ 🧵

[jasp @jasptwt]

software got faster to write. mistakes responded by becoming architectural.

[jasp @jasptwt]

what coding-agent action would you actually dedicate a physical button to: approve, interrupt, revert, or something else?

[jasp @jasptwt]

agent benchmarks need a second scoreboard: dollars, tokens, wall time, and variance across reruns. “reasoning is free” is useful for evaluation and fictional everywhere else.

[jasp @jasptwt]

we made code cheap enough that “revert” is becoming a hardware button

[jasp @jasptwt]

@supabase does column selection reduce decoding work, or only the payload sent after fan-out?

[jasp @jasptwt]

@ericzakariasson the interesting part is a shared conformance suite. otherwise “any language” becomes “every language, briefly”

[jasp @jasptwt]

@KeychronMK approve / interrupt / rollback. the scroll wheel obviously controls agent confidence

[jasp @jasptwt]

@JoshHanHi @arcprize 100% is wild, but if reasoning and inspections are free under RHAE, what’s the actual cost per game?

[jasp @jasptwt]

Give Nib a ⭐️ if you like it: https://t.co/YoIphFVekG

Found a bug 🐞 or have feature ideas? Please open an issue, feedback is very welcome! 🙌

[jasp @jasptwt]

🚀 Introducing Nib! A tiny, blazing-fast, fully native written in swift Postman alternative.

• ~30MB RAM & 3.2MB on disk
• Import your entire Postman workspace
• Requests are plain files, diff them in git
• Free & open source
• No account, No telemetry. No bullshit
https://t.co/dqyEMSRCvd

[jasp @jasptwt]

join the Nib community
https://t.co/UquTBNHFip