25-04#

Chawan#

chawan: TUI Web Browser via chawan: TUI Web Browser | Lobsters

curl https://nim-lang.org/choosenim/init.sh -sSf | sh
sudo apt install libssh2-1-dev libcurl4-openssl-dev pkg-config make ncurses-base

# add ~/.nimble/bin to PATH
v ~/.bash/4-PATH
exec $SHELL

git clone https://git.sr.ht/~bptato/chawan && cd chawan
make
sudo make install

cha -V
mancha cha

  • useful key bindings

    • jump to link prev/next: [, ]

    • history back/foward: ,, .

TrueSkill2#

https://www.microsoft.com/en-us/research/publication/trueskill-2-improved-bayesian-skill-rating-system/

Online multiplayer games, such as Gears of War and Halo, use skill-based matchmaking to give players fair and enjoyable matches. They depend on a skill rating system to infer accurate player skills from historical data.

Sure! As a player I want to be matched with / against people at my skill level.

This extension, which we call TrueSkill2, is shown to significantly improve the accuracy of skill ratings computed from Halo 5 matches. TrueSkill2 predicts historical match outcomes with 68% accuracy, compared to 52% accuracy for TrueSkill.

So TrueSkill 1 was only 2% better than a coin throw. 😅

lvmsync, Matt Palmer#

mpalmer/lvmsync: Synchronise LVM LVs across a network by sending only snapshotted changes via mpalmer/scrape-ct-log: CLI tool to quickly and efficiently bulk-download entries from a Certificate Transparency log and Matt’s Blog: Brane Dump

Coding Assistants#

Today (2025-04-17), OpenAI released Codex (via lobsters).

Some guy released Plandex (via hn).

I’ll give both a go and compare them to Aider.

sqlsync#

Carl Sverre is hard at work on sqlsync / graft:

Git auf Deutsch#

Mach ein Ziehbegehren, wenn du mit der Vereinigung fertig bist!

danielauener/git-auf-deutsch 😅

rust –> wasm –> js#

Rust via wasm-pack into wasm via wasm-bindgen into JS API.

Source: https://youtu.be/oLYda9jmNpk?t=900

CaMeL#

  • paper

  • simonw

  • hn

  • simon’s approach: two LLMs

    • quarantined LLM (Q-LLM) reads possibly malicious inputs

    • privileged LLM (P-LLM)

    • P-LLM only works with references (e.g. $email-summary-1)

    • in essence: P-LLM creates control flow (aka. the program)

  • CaMeL’s security engineering:

    • sandbox Python subset

    • provenance (e.g. send_email(recipient=address): only if recipient is trusted)

    • capabilities and policies

Getting forked by Microsoft#

Read Getting forked by Microsoft | Hacker News if you want to see some Microsoft-hate. The product itself (spegel, a stateless cluster local OCI registry mirror) looks interesting. MS added artifact streaming support in peerd.

On this page