Today’s thread is AI moving from novelty to infrastructure and governance: OpenAI’s claimed math and theoretical computer science breakthroughs led the discussion, alongside debates over whether model weights should be treated like cryptography and where AI actually helps in software work versus where engineers still have to carry products to production. Trust and control run through the rest of the slate too, from Cursor removing cost visibility to Canada’s signing of the UN cybercrime convention and fears of broader surveillance. Also in view: Go’s evolving standard library, a ripgrep musl segfault, worsening NAS design, exposed water-sector PLCs, and treating CI/CD as production.
-
Ten advances in mathematics and theoretical computer science
-
openai.com
- 280 comments
OpenAI posted ten new math and theoretical computer science results it says were found by an internal reasoning model, including proofs and counterexamples for long-open problems, with Lean formalizations for some of them. The comments mostly accepted that several of the targets are genuinely hard, but pushed hard on missing methodology, unclear total cost, and what exactly should count as the real scientific contribution here.
-
-
A Surveillance Treaty in Disguise: Canada Signs UN Cybercrime Convention
-
michaelgeist.ca
- 142 comments
A Canadian privacy-law scholar argues that Canada quietly signed the UN Cybercrime Convention, a treaty sold as anti-cybercrime cooperation but broad enough to expand cross-border surveillance and speech policing. Commenters largely treated it as another example of governments using crime and child-safety rhetoric to normalize identity checks, data sharing, and more state access to online activity.
-
-
Cursor removed cost information from the usage page and CSV export
-
forum.cursor.com
- 127 comments
A Cursor forum post said the company removed dollar-cost information from its usage page and CSV export, making it harder for users to track what model choices actually cost. The comments treat it as a trust and procurement problem, not a UI nit, with a Cursor employee saying the CSV break was accidental but the graph removal was deliberate because plan usage shown in dollars confused users.
-
-
Twenty-five years ago it was cryptography, today it's model weights
-
weeraman.com
- 147 comments
A blog post argues that today’s push to restrict AI model weights looks a lot like past attempts to control cryptography exports: once the code exists, trying to bottle it up mostly harms legitimate users while copies spread anyway. Commenters mostly agreed with the analogy’s broad direction but said the bigger fight is over what is actually dangerous: the weights themselves, the systems built around them, or the political and commercial power of the labs pushing for restrictions.
-
-
AI doesn't generate working products, that's still your job
-
- 261 comments
A blog post argues that large language models are great at getting software to a first demo, but still weak at turning that prototype into a maintainable, production-grade product. Comments mostly agreed from direct experience, with the useful split being where AI works well today: scaffolding, internal tools, migrations, and tightly guided implementation, not architecture or long-lived codebases.
-
-
Golang proposal: container/: generic collection types
-
github.com
- 175 comments
A new Go proposal would add generic collection types to the standard library under `container/`, including common data structures now made possible by generics and iterators. The reaction was mostly less about the specific API and more about whether Go is sensibly maturing or slowly giving up the simplicity that made people choose it in the first place.
-
-
RipGrep musl binaries occasionally segfault during very-large searches
-
- 162 comments
A ripgrep issue report showed that the statically linked musl build can occasionally segfault during huge searches, and commenters quickly converged on this looking like a Linux virtual memory bug that musl happens to trigger more easily. The side story was just as loud: a long AI-generated analysis got mocked as unreadable and likely overconfident, even though kernel developers still found a plausible bug worth chasing.
-
-
Ten Ways NAS Is Getting Enshitified
-
nascompares.com
- 167 comments
A blog post argued that consumer and prosumer NAS boxes are getting worse through soldered RAM, fixed boot storage, vendor lock-in, weaker upgrade paths, and copycat software. The comments mostly agreed with the broad decline, but sharpened the real split between people who want a simple low-power appliance and people using a NAS as their all-in-one home server.
-
-
CISA Alert: Water Sector PLC Targeting
-
censys.com
- 41 comments
A Censys post amplified a CISA alert about attackers targeting water-sector PLCs, noting thousands of internet-exposed Rockwell Automation devices in North America. The reaction focused less on the specific threat actor and more on how routine it still is for water and industrial control gear to be reachable online with weak or default security.
-
-
The development pipeline is a production system
-
sundry.jerryorr.com
- 81 comments
A blog post argues that your build, test, and deployment setup should be treated like a real production system because when it breaks, engineers stop shipping software. Commenters largely agreed on the operational point, then pushed it into practice: separate CI from deploy, mirror and pin dependencies, and decide which internal environments deserve true production-grade uptime.
-
-
NetBSD 11.0
-
blog.netbsd.org
- 83 comments
NetBSD 11.0 is a new release of the long-running BSD operating system, with updates to its firewall, hardware support, toolchain, and a tiny x86 microVM kernel that can boot in about 10 ms. The comments turned into a status check on the BSDs more broadly, with people arguing that NetBSD still stands out for portability, clarity, and old-hardware support even as Linux dominates mainstream use.
-
-
Google has abandoned Google News?
-
elgan.com
- 172 comments
A blog post argues that Google’s news search has effectively degraded, especially around date, location, and source filters, making it unreliable for finding recent reporting on specific topics. The comments mostly agreed that Google search and adjacent products are getting worse, but a sharp subthread challenged the article’s evidence and said the screenshot likely came from Search’s News tab, not the standalone Google News product.
-
-
GitHub has alternatives, but no replacement
-
lalitm.com
- 120 comments
A blog post argues that while GitHub has plenty of technical alternatives for hosting Git repositories, nothing truly replaces its mix of open source community, discovery, contribution workflow, and free infrastructure. Comments mostly agreed that the hard part is not Git hosting itself but GitHub’s network effects, especially free CI and the low-friction path for outside contributors.
-
-
Postmortem for Kernel Soundness Bug #14576
-
leodemoura.github.io
- 38 comments
Lean maintainers published a postmortem for a kernel soundness bug that let an AI-assisted fake “disproof” of the Collatz conjecture pass the verifier. The bug was quickly reduced to a tiny proof of false and fixed, and the comments mostly treated it as a reminder that formal verification is only as strong as the proof checker underneath it.
-
-
A week in Matrix
-
piegames.de
- 67 comments
A blog post from a Matrix admin describes a week of breakage and user pain on a self-hosted Matrix setup, from client confusion to decryption failures and server dead ends. The comments mostly agree the UX and operations story is still rough, while arguing Matrix remains one of the only serious options for federated, encrypted group chat at scale.
-
-
Software for One
-
ajwaxman.com
- 249 comments
A blog post argues that modern AI coding tools make it practical to build tiny custom apps just for yourself or your family, instead of waiting for the app store or SaaS market to care about your niche need. Commenters mostly agreed the real unlock is not shipping startups faster, but turning one-off annoyances, workflows, and personal preferences into software that is finally cheap enough to make.
-
-
Seedance 2.5
-
seed.bytedance.com
- 41 comments
ByteDance posted Seedance 2.5, a text-to-video model pitched around one-take cinematic clips, reference-based consistency, and ad-style production workflows. Commenters were struck by how close the samples look to polished social media and commercial video, but the discussion quickly turned to access, control, and whether closed high-end models will lose ground to cheaper open alternatives.
-
-
Run Kimi K3 using 29 GB of RAM at 0.50 tok/s
-
- 158 comments
A new GitHub project called WASTE claims it can run the very large Kimi K3 language model on a machine with just 29 GB of RAM by streaming most of the model from SSD, reaching about 0.5 tokens per second. Readers found the engineering trick interesting as a proof of what is possible on commodity hardware, but most saw the speed and economics as too weak for normal interactive use.
-
-
BMW Spider-Man in-car advertising
-
consumerrights.wiki
- 158 comments
A Consumer Rights Wiki post documented BMW pushing a Spider-Man promotional animation to some cars’ dashboard screens, reviving complaints about ads and surprise content appearing in products people already bought. The reaction was overwhelmingly hostile, with most readers treating it as another sign that newer cars are becoming attention-hungry, software-controlled platforms instead of straightforward vehicles.
-
-
The Silicon Valley Founder Meat Grinder
-
zaksa.zip
- 131 comments
A blog post sketches the rise and collapse of “Jim,” a startup striver who talked his way into Silicon Valley opportunities, briefly got rich, then spiraled through status chasing, drugs, and bad spending. Commenters mostly treated it as a recognizable type rather than a one-off, and used it to talk about startup culture rewarding performance, risk appetite, and social signaling more than durable craft.
-
-
EU will mandate labels on authentic-looking AI content starting August 2
-
engadget.com
- 104 comments
An Engadget piece says the EU’s AI Act will start requiring labels for realistic AI-generated or AI-manipulated media, especially deepfakes, from August 2. The comments mostly agreed with labeling deceptive content, but focused on narrower questions: what exactly counts as “authentic-looking,” whether labels will become useless boilerplate, and whether provenance systems would help more than disclosure rules alone.
-
-
Diátaxis
-
diataxis.fr
- 22 comments
Diátaxis is a framework for structuring technical documentation into four types such as tutorials, how-to guides, reference, and explanation. The comments are largely enthusiastic from people who have used it in real doc rewrites, with the sharpest pushback aimed at treating it as a rigid doctrine rather than a useful organizing lens.
-
-
June in Servo: real world compat, media queries, SharedWorker, and more
-
servo.org
- 61 comments
Servo, the Rust-based browser engine project spun out of Mozilla, posted its June progress report covering better real-world website compatibility, media queries, SharedWorker support, and ongoing work toward being usable as an embeddable engine. The comments were broadly supportive, but the useful signal was that most people still see Servo as promising infrastructure rather than a browser you can deploy widely today.
-
-
Solid Queue 1.6.0 now supports fiber workers
-
- 38 comments
Rails job queue Solid Queue 1.6.0 added fiber-based workers, letting one worker process many I/O-heavy jobs without a thread per job. The comments mostly saw this as a practical win for common background-task workloads, with caveats around CPU-bound jobs, blocking libraries, and benchmark hype.
-
-
Explorative modeling: Train on the best of K guesses
-
alexiglad.github.io
- 21 comments
A blog post and paper propose “explorative modeling,” a training trick where a generative model makes K guesses for each example and only learns from the best one. Readers saw the reported image results as potentially big if they scale, but much of the value in the comments was about placing it in older work on latent-variable models, winner-take-all training, and diffusion pipelines rather than treating it as a brand-new paradigm.
-
-
The Art of 64-bit Assembly
-
nostarch.com
- 84 comments
A new 800-page No Starch book on x86-64 assembly, focused on Windows and MASM, sparked less discussion about the content than about its publisher copy, tool choices, and whether LLMs make assembly books obsolete. The useful signal was that longtime readers still rate Randall Hyde’s low-level teaching highly, but the book is clearly aimed at people who want deep platform-specific understanding, not a general cross-platform assembly reference.
-
-
Pgtestdb's template cloning approach to testing is fast
-
brandur.org
- 36 comments
Brandur benchmarked pgtestdb, a Postgres test helper that creates clean databases by cloning a pre-migrated template instead of rerunning migrations for every test. The comments mostly agreed the pattern is practical and fast, then compared it with transaction rollbacks, shared “dirty DB” testing, and table-clearing tricks that can be even faster in some setups.
-
-
Long Range Wi-Fi – Pushing 2.4 GHz Wi-Fi to the limits (2019)
-
phidgets.com
- 61 comments
A 2019 Phidgets writeup tested how far ordinary 2.4 GHz Wi‑Fi could be pushed with directional antennas and unlicensed power limits. The comments say the result is more nostalgia than breakthrough, then add where the real limits come from today: protocol timing, spectrum rules, line of sight, and the fact that cheap point-to-point gear already does multi-kilometer links.
-
-
The Absurdity of Albert Camus
-
historytoday.com
- 87 comments
A History Today essay on Albert Camus prompted a wide-ranging discussion of what people actually get from his most famous works, especially The Myth of Sisyphus, The Stranger, and The Plague. The strongest takeaway was that many readers now see early Camus as only the setup, with his later work giving the more usable answer: do your job, act decently, and keep going without pretending the universe will supply meaning.
-
-
How to Exist
-
raptitude.com
- 228 comments
A Raptitude essay argues that many people find it surprisingly hard to sit still and do nothing for even a few minutes, and that practicing this can loosen anxiety, rumination, and compulsive busyness. The comments mostly turned it into a blunt reality check on meditation: useful for some, empty for others, and often confused with a much broader set of practices and life conditions.
-