GitHub's new Git storage architecture
A commenter notes that GitHub's new architecture is similar to what enroute.sh did to scale Git.
30 stories
Get a notification every evening
You hear about the digest when it is ready. Turn it off any time.
Notifications are on
You will hear about each evening’s digest when it is ready.
To get notifications on iPhone or iPad, add this site to your Home Screen first: tap Share, then Add to Home Screen. Open it from there to turn notifications on.
Notifications are blocked in this browser. Allow them in site settings and reload.
New notification sign-ups are closed for now. You can follow along with RSS.
Notifications could not be turned on. Check your connection and try again.
Notifications could not be turned off and are still on. Check your connection and try again.
The day’s top discussions
A commenter notes that GitHub's new architecture is similar to what enroute.sh did to scale Git.
A commenter claims that using the Qwen model can yield nearly equal quality answers from different Claude models (Haiku, Sonnet, Opus).
A commenter questions the claim that Kugelblitz runs smoothly even on M1, recalling that Google Earth also ran smoothly 25 years ago.
A commenter says they used to wish for OSX on other hardware, but now they want anything but macOS on Mac hardware.
A commenter appreciates that Mitchell refrained from including his own competing product Rex when listing tools built on Libghostty.
A commenter is skeptical about relying on Shodan data, stating from their experience that most online devices are already honeypots.
One commenter questions Replit's ecosystem while another compares their work to OTP supervision trees and process isolation.
One commenter praises AI as a research assistant for plotting and literature search, while another criticizes the report for lacking comparison with non-AI users.
Commenters point to sample videos on the product page and request quality+time comparisons between different quantizations.
Commenters compare the project to Nix and Bazel's architecture pattern, and question the idea of a WASM-based durable execution engine.
One commenter says Memorilla injects "vibe" into LLMs token-efficiently, good on concepts but less good on quotes/exact text retrieval.
A commenter wonders why parallel Top-K aggregation per partition wasn't chosen, offering both reduction and correct maximum for pruning.
A commenter says at first "I can build this myself", but after browsing the site realized the value proposition.
A commenter explains buffering should use work-or-time based flushing, with time-based flushing critical for latency bounds in low-throughput environments.
A commenter points out Tata says the issue has been fixed and no leaks occurred.
A commenter says it's built on tmux-continuum & tmux-resurrect, used for years, eliminating worries about losing panels and scrollback.
A commenter notes the article defines memory as prediction of future needs but sidesteps continuous learning, suggesting a different research direction.
The discussion revolves around the importance of shifting audio/video sync during playback and VLC's lack of support for negative timestamps.
Commenters note the ID requirement is for sanctions compliance and reducing low-quality submissions, while another argues a misunderstanding of OAI and MCP's guarantees.
A commenter notes that open webui and open terminal were mentioned in the thread but not in the article.
A smaller, more technical community
A commenter says merging musl and Nvidia drivers is easier than making a responsive web page on mobile (they are on iOS).
Commenters suggest adding Servo to resources like MDN or caniuse, and note the browser works on some news sites.
Commenters discuss alternatives like ignoring errors or using simple assertions, while the author argues MustCreate better expresses logic unrelated to the system under test.
The week’s most-starred new repositories