Mistral Large 4
Commenters note that Mistral is still competitive and strong in vision and cybersecurity benchmarks.
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
Commenters note that Mistral is still competitive and strong in vision and cybersecurity benchmarks.
One commenter warns about interpreting benchmarks carefully, while another finds Polars better than pandas.
Commenters argue that OpenAI is responsible for its agents' actions and should be punished.
Commenters are skeptical that derivative-free optimization methods will be impactful.
Commenters are surprised at the ease of writing game logic in SQL, while one calls it engineering malpractice.
Commenters suggest alternative tools for elevation and grade information on bike routes.
Commenters note that users have limited options against data sharing in connected vehicles.
One commenter notes Jev outperformed LLM judges on citation support tests, while another argues Jev excels in multi-step reasoning.
One commenter suggests using C++ instead of macro-based hacks, while another mentions writing an article on type-safe containers in C23.
One commenter finds the magnet introduction bizarre; another sarcastically claims their own agents discovered cold fusion.
One commenter refutes the claim, emphasizing the need for human prompting; another jokes about metal skeletons.
A commenter notes these aren't confirmed compiler bugs yet, pointing to a similar OpenSSL inline asm issue and possible undefined behavior in quiche.
Commenters argue the 'natural equilibrium' concept is a 1950s cybernetic machine fantasy, citing the North American cod fishery that never recovered.
Discussion revolves around whether PowerPC-style hashed page tables are better than tree-based ones and whether the right solution has changed since 2003 as scales grew.
Commenters note that since GLM 5.2, reverse engineering closed protocols from binaries has become much easier, with some models even instrumenting running binaries automatically.
Commenters express a wish to remove unused simulator images and a return to Linux due to Apple's bloat and lock-down.
One commenter calls Muse clickbait and defends its VM isolation, while another says they don't trust agents with bank/email access.
Discussion focuses on the importance of methods like confidence intervals, comparisons, and noise balancing in benchmarking.
A smaller, more technical community
The user shares their experience hosting email with maddy and iOS client delays; a commenter points to a similar discussion thread.
One commenter asks about running Factor with Musl Libc on Alpine, while another shares positive experience, calling it fast and ambitious.
Discussion revolves around open source becoming harder for maintainers but AI agents benefiting users, and how the GPL addresses this.
Commenters debate trust in software in the vibe coding era: one worries about backdoors, another argues LLMs make auditing easier.
The week’s most-starred new repositories