NanoraptorReviewers noted that while Vim for Gameboy featured one of the deepest command systems on the platform, its soundtrack consisted primarily prolonged silence punctuated with occasional terrifying beeps.
| Github | https://github.com/mikesiegel |
Mike Siegel
- 98 Followers
- 143 Following
- 724 Posts
I do various offsec stuff.
Joe Grand
At some point the frontier AI companies decided "Let's train a model with the entire fucking internet and every piece of YA fiction in our quest for AGI." Then they ascribed person-hood to their model, as can be seen in their system cards when they have psychologists interview it. Then they sell you this service, which has the uptime of a potato to do serious work.
Why aren't these companies asking themselves: "Does my vulnerability finding model benefit from knowing all of Taylor Swift's discography and the entire content of 4chan?" or "Why does my status board look like a Christmas tree?"
Models that are 300 times smaller can produce similar results at a fraction of the cost, on your desk instead of a data-center.
Sadly, it's hard to run them on your desk given GPUs are so expensive from someone trying to birth a sentient PyTorch model.
Michel Lind 
RE: https://hachyderm.io/@selfcare/116815727262141871
So this occurred to me to try a couple of days ago, and what a timely post from @selfcare here ... I'm trying out #NoAIFriday - and likely extending it to the weekend too.
You might be forced or feel pressured to use #AI - specifically #LLMs - at work. Your community projects might be #AIpilled as well. But apart from the ethical and environmental reasons why the current LLM proliferation is both bad but also unsustainable -- it's also bad for you personally! It can cause burnout (HBR: https://hbr.org/2026/03/when-using-ai-leads-to-brain-fry); it can cause #CognitiveImpairment (MIT, Cornell via BBC: https://www.bbc.com/news/articles/cd6xz12j6pzo)
So I'm practicing #SelfCare by going cold turkey on Fridays - artisanal coding only (I'll make an exception for #OncallDuty) followed by a mild form of #TechShabbat (https://en.wikipedia.org/wiki/Technology_Shabbat) over the weekend
daniel:// stenberg://I reported a possible security problem against OpenSSL 4.0.1 found with no AI at all. Like a cave man.
Royce WilliamsRE: https://mstdn.social/@TechCrunch/116807118492991232
This is incorrect. Physical security keys are more secure than passkeys. All of the strong phishing resistance, none of the software-brokered attack surface.
To be clear, passkeys are safer than push or SMS or email or passwords. And physical keys are a pain to maintain for the same reason they are more secure: you can't synchronize their secrets anywhere.
Kevin BeaumontVery funny screw up by Meta. They installed a keylogger on employees PCs, so they could train their AI. However they exposed the collected data online, so anybody could see what anybody who works there has been typing.
https://www.wired.com/story/meta-accidentally-let-employees-access-each-others-keystroke-data/
Joe BrockmeierRE: https://social.pincade.com.au/@Dozer/116793074415724273
Oh, wow. Happy anniversary Quake. Many happy hours playing this one in my twenties. It really was a big step forward visually, and the gameplay was great. Non-stop pedal-to-the-metal action.
Jake WilliamsApple fumbled the implementation of its USB controller on A12 and A13 devices so it's possible to compromise the boot loader. Exploit code has been published. This can't be patched with a software update.
Lukasz OlejnikChinese Qwen3.6-35B-A3B model with only ~3B active parameters, running on a Mac laptop produced finding sets comparable in size to frontier cloud models (GLM-5, Claude Opus 4.6). https://srlabs.de/blog/beyond-fable
