OpenAI touted GPT-4's scores on professional exams and other standardized tests. But they may have tested on the training data: we found slam-dunk evidence that GPT-4 memorizes coding problems that it's seen. Besides, exams don't tell us about real-world utility: It’s not like a lawyer’s job is to answer bar exam questions all day.
The latest in the AI Snake Oil book blog by @sayashk and me: "GPT-4 and professional benchmarks: the wrong answer to the wrong question" https://aisnakeoil.substack.com/p/gpt-4-and-professional-benchmarks
#Twitter is having a mysterious bug: some tweets are not showing up in the feed, and and are shown as "not exist" and "deleted" if viewed via a link from some regions, for no apparent reason.
Some users are so prolific and so high-quality uploaders that just visiting their "user uploads" page is an exquisite curation it itself.
Such it is with THE IBIS REBELLION, who has been scanning in and uploading discard pulp magazines with a nod to quality and evocation for years:
#Swift folks, we're busy working on a macros for the Swift language and would love your thoughts. It's a big feature with a lot of details that need to be right. We started by laying out our vision in a macros vision document, talking through the high-level approach to introducing macros: https://github.com/DougGregor/swift-evolution/blob/macros-vision/visions/macros.md
It’s time to shine some light on the connection between recent “anti-groomer” rhetoric (targeting our LGBTQ friends and neighbors) and established white supremacy initiatives.
In December 2022, on telegram, the white lives matter (WLM) initiative promised a coming revision of their “activist” manual. This is the PDF manual they distribute to recruit and guide white supremacists in promoting hate through banner drops, sticker campaigns, etc.
So far a WLM revised manual has not appeared, but in January 2023 another “initiative” appeared on the same online platform calling itself “Project 171” or “Anti Groomer Action.”
CVA researchers have determined Project 171 is an attempt by WLM to pivot tactics away from hard white supremacy rhetoric and instead, capitalize on the recent wave of hatred targeting schools and drags shows with accusations of “sexualizing children.”
The following comparisons of the WLM activist manual 2.0, and the January 2023 Project 171 Anti Groomer Action manual show identical format, organization, and in many cases, cut and pasted text from one to the other.
We don’t intend to reprint the entirety of either manual, but our comparisons provide evidence that the recent Project 171 initiative is an attempt by an established white supremacy organization to diversify and infiltrate a parallel hate narrative targeting LGBTQ people.
So it turns out Rust supports incremental compilation and multiple codegen units, and it just wasn't hooked up to the kernel build system!
I hacked that in and now trivial driver changes take 4-5 seconds to build instead of 30~ ✨✨
https://github.com/AsahiLinux/linux/commit/6cb6d0b4fbbe5d99e82829e9c20618f85b5d890a
I find it interesting to see, by a kind of revealed preference, what people really value via what qualities about someone they criticize or insult.
E.g., in the progressive circles I run in everybody agrees that body shaming is wrong, but shaming people for having having body parts or disabilities that don't match the standard masculine ideal is their go-to default insult, so it's pretty clear they don't mean it when they say all bodies are ok or even that trans men are real men.
I should caution, this is a *preliminary* result, as doing real performance analysis with solid methodology is hard. But it is exciting nonetheless.
I have the paris-30k demo running at 120fps on a mac M1 Max. This is with some tweaks and hacks, but I believe all that can be applied in production, it'll just take some work.
I'm really looking forward to doing careful benchmarking, as I believe Vello will come out way ahead of competitive renderers.
Cf. my recent boost, the Total Replay collection of Apple II arcade games is verrrry good. Super easy to use in #AppleII #emulation.
As a slightly unusual demonstration: I've been noodling with it on my #SteamDeck. Details on cohost: https://cohost.org/joel-k-baxter/post/839091-apple-on-deck
There are a lot of private space companies who still use “manned” instead of “human” or “crewed”.
#nasa hasn’t used “manned” in our official nomenclature for ~ 20 years. It may sound pedantic, but women are still only 20% of engineering graduates - the same percentage as when I graduated.
Language and representation - think of job postings - matter.
#space #engineering #WomenInSTEM
https://www.theatlantic.com/science/archive/2019/07/manned-spaceflight-nasa/594835/
あけましておめでとうございマストドン
my year 9 computer science teacher introduced me to Processing and to The Coding Train. Honestly I had so much fun working in this language.
If you have a casual interest in computer graphics and art, processing is such a fun language - it's pretty accessible too
The coding train is just an awesome YouTube channel, he does loads of programming challenges and explains lots of computer science concepts in a really accessible way.
"the bob ross of computer science" ☺️
Anyone have other examples of "Here's a common thing that most people are doing wrong, along the lines of https://blog.ganssle.io/articles/2019/11/utcnow.html and https://blog.ganssle.io/articles/2018/03/pytz-fastest-footgun.html ?