Herd immunity : not just for humans. Your IT security team could be preventing your computer from becoming impervious to cyber-threats. Delete your antivirus!

@ChristosArgyrop I have also been wondering about word size and the trend in machine learning to reduce it. A multiplier circuit's gate count is O(N²), right? So in a sense, 2x the parameters at 1/2 precision halves algorithmic intensity. Why is that generally a worthwhile tradeoff? Why doesn't […]

#vibecoding gems

Lol, this #vibecoding advice is the stuff that CVEs are made from

Based

I have no idea which of the 8 disks RHEL decided to install itself at. #fafo

This is where it gets interesting : at a word size of 1 bit, multiplication is just logical and. So is all multiplication just logical conjunction? Whereas the owner of the site is unlikely to be aware of the connection, first principle derivation of […] [Original post on mast.hpc.social]

Nailed it

It is insane, that #intel kneecapped i7's for scientific work. The 11th gen i7 offered the AVX512 instruction set which DOUBLED throughput compared to the 10th 7 the 12th gen. To get the same throughput that this i7 delivers in our RNAseq workflows, one has to pay $$$ for a Xeon

I am honestly impressed by the SIMDe (SIMD everywhere) library https://github.com/simd-everywhere/simde It simplified considerably some #bioinformatics relevant code I am writing since I don't have to reinvent the wheel and use complicated preprocessor directives and macros to figure out where […]

Every single night the quality of my AI generated slop code (Grok through Githug copilot) goes down substantially and reproducibly. Is it because most people use Grok to generate pork after hours, ie the 2020s version of Twitter after dark?

My most successful #perl modules pass the tests with flying colors

I just released theAlien::SIMDe (https://github.com/chrisarg/Alien-SIMDe) , that allows one to include the portable SIMDe intrinsics library (https://github.com/simd-everywhere/simde) in Inline::C modules . The "Parsing Gigabytes of JSON per second" https://arxiv.org/html/1902.08318v7 may be a […]

Profile HMMs and other Hidden Markov Models explained from we @weratedags.com at #bsky #bioinformatics #generativemodels #probability #latentvariables #statistics

One of the rare #perl module tha talks directly to GPU with or without PDL (Perl Data Language) intermediaries. https://github.com/chrisarg/Task-MemManager-Device . I can see the path to a #Perl #gpu ecosystem with PDL playing a major role, courtesy of #openmp.

Can you feel the AGI?

Back to torturing the chatbots tonight

The slopbots didn't fail to code a sys admin utility in Perl tonight. Looking at the stolen code I'd say the original was written with Perl v5.8 but used the PBP book (released in 2005) conventions for documentation. So it was pieced together from stuff written between 2005-2009.

Are there any #gpus offering high counts of integer operations per second? I have tested a bunch of NVIDIAs, and their integer arithmetic throughput is definitely smaller than their FP32.

Teaser : movement to, modification at and movement from, of #Perl Scalars in the GPU, leveraging https://metacpan.org/release/CHRISARG/Task-MemManager-0.08 to mix #PDL and #GPU calculations & coordinate memory mapping CPU ↔️GPU via #OpenMP. Functional code […] [Original post on mast.hpc.social]

This is hard-core and deserves respect https://gaultier.github.io/blog/x11_x64.html#opening-a-socket

My takes as an amateur coder after a "vibecoding" session with #ai in #githubcopilot: 👉 Agentic bots are unhelpful 🤏“Ask” mode could generate useful code, riddled with mistakes. 👍As auto-complete for documentation helping me overcome writer's block […]

AI is going GREAT! https://substack.com/inbox/post/172538377 (apologies for a substack link but it's a good post)

Next level #python delusion just dropped in #twitter. Best used with #ai hallucinations for top notch irreproducible research and useless #insights. #canyoufeeltheagi? #pythonistas

I did a small #vibecoding experiment last week in #perl to interface to a small C library for a bigger project. The mistakes #claude made were rather subtle ; one can not easily pick them them without lots of proofreading Look at the relevant section 👇 https://metacpan.org/pod/Bit::Set

Every #ai #llm #chatbot in #githubcopilot fails to understand the difference between keys and values when #vibecoding the #perl API to a simple #c library. In this case the terms on the two sides of the arrows should be reversed. https://github.com/chrisarg/Bit

Another experimental exhibit that LLMs cannot reason. Setting medical diagnosis (an area in which they are pushed hard). In this experiment they took questions in which the LLMs had previously scored well and changed the correct answer to None Of the Above […] [Original post on mast.hpc.social]

Gemini Pro 2.5 stopped responding to my vibe coding. I broke both of its wolves 🤣🤣🤣🤣

Does the P in GPT stand for Pre-trained or Potty-Trained? #ai #mystery

Dunning-Kruger curve, with and without AI. The era of Automated General Idiocracy is rising From: https://x.com/ryzerth/status/1959447949341057428?t=Xu10DeRtTg14uIHdpmdlDw&s=19

Is the spirit of the linear algebra library running the autoregressive multinomial logistic regression model that was trained in the vector database built after numerous copyright violations in the room with us now?

The promise of #vibecoding or #ai in general is an empty one and as likely to be realized as losing weight without diet, exercise and/or Ozempic: a pipe dream sold by the grifters to the gullible and the immature victims of the culture of make believe

Categories of people who engage in anthropomorphic discourses or fantasies about AI and how to deal with them: 1. Grifters ➡️ call out 2. Mentally ill ➡️ suggest professional help 3. Immature ➡️ ignore 4. Morons ➡️ troll 5. Ignorant ➡️ educate

The 15yo daughter comes back from coding class and they tell me Rust is on the curriculum! What is even going on?

Surely, there must be a cheaper and less energy intensive way to train/estimate an autoregressor on a vector database than an LLM.

Of course outsourcing the low-level parts of digital technology is unavoidable. We cannot print chips in our backyard any more than we can construct printing presses. But the disenfranchisement and widespread lack of #digitalliteracy that now reduces people to buffoons reflects a unique […]

yet the capture, dumbification and manipulation of the digital realm has been going on for decades, with ever increasing audacity, with puny resistance (basically just from the 0.1% or so of #opensource and #fediverse rebels and "fanatics"). This reveals the hollowness of institutions, the […]

The original vision of 60's-70's digital technologists was the #augmentation of human skills (memory, knowledge management etc.) One way or another that vision *has* delivered but the landscape has changed. Today the most urgent need is to protect from tech-powered monsters like […]

Human Friends don't let CPU core friends suffer under #python

"Paying attention to dots leaves no time for products" Anonymous vibe coding founder during an SEC investigation for fraud. #mathofai

#ai has a fundamental economics problem explained by @zoomyzoomm on X. Once the VC money disappears , the house of card will collapse unless the user coughs up the cash. Is this slop or the AI generated porn worth to you one to 2 k a year?

I like strongly typed languages like #assembly : everything is a byte, a word, a double word, or a quad word. All other types are social constructs #programming

Knowing that AI (slop) is as good as it gets for the next 10-15 years (there will likely be an improvement in Automated Inference for sure when the LLM noise dies), do we just accept the slop until then or do we revert back to 2021?

Apparently, #gpt5 (and I assume all the ones prior to it) are trained in datasets that overrepresent #perl. This, along with the terse nature of the language, may explain why the Perl output of the #chatbots is usually good. https://bsky.app/profile/pp0196.bsky.social/post/3lvwkn3fcfk2y #llm #ai

The success stories about #ai in medical diagnosis are not #agi but are due to: 1) Physicians blowing off patient concerns (unfortunately, it's common among our profession to call ppl whose symptoms we can not explain "crazy") 2) upsampling of "zebras" in #llm training datasets

PhD level of intelligence said the grifters and the gullible rejoiced Source https://x.com/colin_fraser/status/1953606476184125572?t=1XLjhv_YIJret5wOxG4CdA&s=19

So, do all #blas functions scale linearly only up to 4-6 threads? This seems to be the case when multithreaded BLAS is used for glm(m) modeling in #rstats #hpc

RE: failure of Units of Study to promote literacy among US children. The essence of the method "children were expected to figure out the word from the first letter, context clues, or nearby illustrations" is akin to how #ai is trained Predictable #fafo […]

Interesting convo with @jimsalter whether it would be possible to nuke an LLM chatbot with a zip/fork bomb. Consider contributing if you are into hacking because the question is interesting both academically and from infosec perspectives. https://fosstodon.org/@jimsalter/114972081647169022

How to deal with intrusive #ai 👇🖕