← Feed

Posts

• Post #4196033

  It's 4am and all's quiet

• Post #4196032

  "Statistics show that emails are hardly read these days" My percentage of opens may have gone down, but that's only because the amount of spam has gone up. The number of actually meaningful emails I read probably hasn't changed.

• Post #4168943

  RE: https://flipboard.com/@xdaofficial/xda-developers-ntraf9plz/-/a-1Hhvp79zRZ2qM0QiQ8s1cQ%3Aa%3A2772481191-%2F0 Old software didn't change, it just looks better over time because so much new software is all crap.

• Post #4108137

  Region-locked cars are a thing now. Ohjoy. https://carnewschina.com/2026/07/23/zeekr-9x-owner-stranded-abroad-remote-lockout-disables-smart-features-for-30-hours/

• Post #3921871

  Tune LMDB, Entry Cache and DN Cache in 389 Directory Server Tune 389 Directory Server LMDB map size, entry cache, DN cache, and normalized DN cache using dsconf backend config, monitor dbmon, and memory budgeting on MDB backends. https://www.golinuxcloud.com/389-directory-server-lmdb-tuning/ SMH. I wrote #LMDB so that we could eliminate the entry and DN caches in OpenLDAP. Once again, the 389DS guys are Doing It Wrong. They've perpetuated the cache tuning headaches of BDB instead of prope...

• Post #3915157

  Recycled thoughts: physical networks are inherently natural monopolies and should just be acknowledged and operated as such. Electricity, water, telecoms... their usefulness grows with their scale. They should be nationalized, community owned and operated. VNOs and fake competition all selling service on the same physical infrastructure is just a wasteful bureaucratic fiction. Eliminate them.

• Post #3867240

  Reminder: AI "generated" code is 100% plagiarized. You must not accept code of unknown provenance into your code base. Doing so opens you up to potential copyright infringement lawsuits. Nobody needs a repeat of the SCO vs IBM lawsuit over ownership of Unix. Accepting AI-assisted code is just legally untenable. That's black and white, there's nothing to debate. Projects that accept it are idiots and should be shunned. https://mastodon.social/@hyc/114777864519941643

• Post #3790264

  RE: https://hachyderm.io/@badnetmask/116913005775228849 Tech debt being created "just because I can", not because you should, struck a chord. I think it's just the latest incarnation of people who got into programming because they heard you could make money at it, and not because they had any real interest in the technology.

• Post #3546892

  An overview of Virtual Memory (VM), a fundamental concept in modern computing https://dev.to/sangyog2058/csapp-chapter-9-virtual-memory-deep-reference-e46

• Post #2938719

  Past noon, solar panels at over 100% of rated capacity

• Post #2938718

  Hallo-Live: Real-Time Streaming Joint Audio-Video Avatar Generation The method adopts a causal dual-stream DiT model to generate synchronized avatar video and speech in a streaming manner. Hallo-Live reaches 20.38 FPS with 0.94 s latency on two NVIDIA H200 GPUs, while preserving strong lip-sync accuracy, visual fidelity, and speech quality. Built on #LMDB https://github.com/fudan-generative-vision/Hallo-Live

• Post #2575760

  "389-DS (Red Hat Directory Server) is the production choice for >1M entries — purpose-built for large directories with a dedicated replication engine" https://linuxcent.com/tag/openldap/ Pure BS. #OpenLDAP serves millions of queries per second with DBs with over a billion users. 389DS doesn't even manage 1/10th of that. 389DS's /RedHatDS's underlying tech was so outdated they had to abandon it all and adopt OpenLDAP's libraries just to s...

• Post #2322852

  This might become the first mass produced solid state battery to reach the market. https://carnewschina.com/2026/04/14/gac-backed-company-targets-worlds-first-mass-producible-solid-state-battery-in-2026-report-says/ "Energy density: Single-cell energy density reaches 260-500 Wh/kg, higher than traditional liquid lithium batteries, improving range while creating space for lightweight and integrated vehicle design." That's a ridiculously good figure, far better than cur...

• Post #1706744

  It's easy enough for the #OpenLDAP Project to progress without using AI. All of that stuff is just regurgitated stackoverflow anyway, and we've never copied code from there either. Anything we need to know how to write was documented in an OS manpage or ISO or IEEE or POSIX spec. When you RTFM you don't need stackoverflow or chatgpt or claude code.

• Post #1171578

  I find it weird how people only trust projects with lots of commit activity and merge requests. In mature projects a high commit activity would be embarrassing to me, because it shows that a lot of mistakes were still being made. Eventually a project should be "done". And stay done.

• Post #1121346

  From trees to graphs: speeding up vector search 10x with Hannoy - on #LMDB https://blog.kerollmops.com/from-trees-to-graphs-speeding-up-vector-search-10x-with-hannoy

• Post #1121345

  This was a trivial enhancement that should have been committed a while ago: adding support for 2-phase commit to #LMDB https://git.openldap.org/openldap/openldap/-/merge_requests/811 It's another feature that BerkeleyDB has had for a long time, used to implement distributed transactions. Of course BerkeleyDB has full blown distributed access support, and its own Tuxedo-compatible distributed transaction manager. LMDB remains single-node only; the calling app will have to deal with actua...

• Post #861123

  Despite all the problems associated with using AI, I still see some people saying "yeah but if you're careful and vet its output it can still help you". I'm reminded of the kid digging in a giant pile of manure "with all this horse shit there must be a pony around here somewhere!"

• Post #774316

  RandomX version 2.0 released https://github.com/tevador/RandomX/releases/tag/v2.0 Changes: Program size increased from 256 to 384 CFROUND tweaked to change the rounding 16 times less often Group F and E register mix is done using 16 AES operations (the extra 262144 AES ops come from this change) Dataset prefetch is 2 iterations ahead instead of 1 v1 hash: 4194304 VM instructions + 262144 AES = 4456448 total ops v2 hash: 6291456 VM instructions + 524288 AES = 6815744 total ops...

• Post #774308

  RE: https://hachyderm.io/@petrillic/116290913314692206 I haven't written a syntax error in at least 35 years.

• Post #774307

  Bigger isn't always better in AI. Sometimes, a 23-Million parameter model can absolutely crush a 1-Billion parameter giant—if you know exactly how to engineer it. - Packed my dataset into a high-speed Lightning Memory-Mapped Database ( #LMDB ) to feed the Vision Transformer at blistering speeds. I effectively built a model that is much smaller (~300MB vs 2.6GB), runs inference in milliseconds instead of seconds, and improved accuracy by over 17%. https://www.linkedin.com/posts/jatin-m...

• Post #705526

  LLM-Driven Discovery of High-Entropy Catalysts via Retrieval-Augmented Generation CO2 reduction requires efficient catalysts, yet materials discovery remains bottlenecked by 10-20 year development cycles requiring deep domain expertise. This paper demonstrates how large language models can assist the catalyst discovery process Data Access: Downloaded from Open Catalyst Project repositories as PyTorch Geometric Data objects and #LMDB files. https://arxiv.org/html/2603.15712v1

• Post #655375

  Pro tip: software is written to solve particular problems. But all proprietary software is written to solve the same problem: how to extract maximal profits from consumers. If it's not open source, it's not worth running. If you're a software professional, and the projects you work on are proprietary, you're part of the problem, not the solution.

• Post #655372

  Another LMDB: lichen metabolites database. I'm sure there's a lichen subscribe pun hidden in there too

• Post #655371

  redb is a pure #rust embedded DB inspired by #LMDB but experienced programmers prefer the original. Convenient juxtaposition of these two search results...

• Post #641581

  It's kind of astounding that people could get so far into computing that they could create LLMs in the first place, while never encountering the phrase "garbage in, garbage out". Like, how is it even possible to not realize that the system you're building is worse than the sum of its inputs?