26-09#
Audacity 4.0.0 Released | Lobsters
How We’re Redesigning Audacity For The Future - YouTube is a great watch
-
sounds interesting, because I don’t love claude code as an interface, but the subscription is cheap
I want to let agents build next to me (like devs) on MRs and stuff, but do not want Github / Copilot vendor lock in
On not becoming a cyborg | Read the Tea Leaves
a write who was alarmed when he thought “load-bearing” to himself
Any Nix package, live in your browser | Farid Zakaria’s Blog
Understanding Revsets for a Better JJ Log Output — Will Richardson via Conversations with JJ | Lobsters
Tutorial for Septabee, a free DAW in the making for over 20,000 hours | Lobsters
ChiPass Release 2026.09.0 | Lobsters
a fork of KeePassXC after KeePassXC started permitting LLM changes
-
a beautiful long-winded way of saying “use Borg Backup”
based on the premise “i can simply implement backup myself”, the author explores the obvious optimization steps
-
Kinderserie
Beyond jj: config & tools ecosystem
jj tug–>jj advancejj arrangelooks nice to clean up history before pushing“high level workflows as aliases” sounds like a good idea
Open Source alternative to Cloudflare Tunnel
mifi/lossless-cut: The swiss army knife of lossless video/audio editing
tarball or AppImage (instead of snap) available here: mifi/lossless-cut
Ju! Ju! Tsu! by Arialdo Martini
see also Jujutsu VCS (jj)
Laya — 33ms Multilingual System 1 Decision Engine with Calibrated Probabilities
Recommendation: Keep choice schemas under 20 options, or use a two-step coarse-to-fine hierarchy.
Laya: open-source alternative to TypeSafe Jev, runs locally - the “marketing page” by brain function collapse
cleder/awesome-python-testing: Collection of awesome 😎️ Python resources for testing
plasma-umass/slipcover: Near Zero-Overhead Python Code Coverage might be interesting
Apache Arrow based data stores, because zero-copy sounds great
-
FDAP (Apache Flight, DataFusion, Arrow, and Parquet) high level components https://youtu.be/I-Z7kFGsYRI?t=448
-
via https://isnichwahr.de/post/43600 kommentar von Nicci
-
For emails, I get 95% accuracy with this method, with only 50-100 examples for training
sounds like fun :)
The Future of Deployment Units#
I believe that software will become more malleable. Imagine a feedback input in your program that directly feeds an agent that in turn changes your program. We might even see branching (divergence) and merging (convergence) for individual users. Individuals might implement a function and other people can get it - like an app store, but on feature granularity.
Logit#
Jev in 25 lines of Python - NobodyWho says:
Massage the logits into probabilities.
But what is a logit? Basically: “even” at 0 and probability between -∞ and ∞.
inverse operation: Sigmoid (from [-∞..∞] to [0..1])
odds = log(p/(1-p))
ranges
p: [0..1]
odds: [0..∞]
logit: [-∞..∞]
even (50/50)
p: 0.5
odds: 1/1 = 1
logit: 0
note: Softmax is Sigmoid for many: take list of logits and ensure they all turn into probabilities that sum up to 100%