Sign in

gwd3.bsky.social

@gwd3.bsky.social
37 followers 42 following 136 posts
PostsRepliesMedia
gwd3.bsky.social @gwd3.bsky.social · 22/08/2026
"You have garlic in your soul, Mr. Grinch!". Your soul is like one of the two powerful aromatics on which western cooking is based, which infuses everything around it with a lovely creamy richness? How is that an insult?
000
gwd3.bsky.social @gwd3.bsky.social · 21/08/2026
Complaining about Claude’s inability to write my design document in plain English, while I tell it, “Can you rewrite this series so the perdomain functionality uses xenheap pages, and walk via maddr-to-vaddr conversions, rather than using domheap pages and walking with an irq-safe map-unmap?”
200
gwd3.bsky.social @gwd3.bsky.social · 18/08/2026
Really wish this could be an Apple Watch face -- so much practical information right there. sunclock.net
sunclock.net
Sun Clock
A 24-hour clock that shows sunrise, sunset, golden hour, and twilight times for your current location. It also shows the current position and phase of the moon, and its rising and setting times.
000
gwd3.bsky.social @gwd3.bsky.social · 16/08/2026
Can’t tell if the ad for mayonnaise on a webpage or a recipe for homemade mayonnaise is stupid or genius.
000
gwd3.bsky.social @gwd3.bsky.social · 11/08/2026
I am having way too much fun with Sonic Pi sonic-pi.net
sonic-pi.net
Sonic Pi - The Live Coding Music Synth for Everyone
Sonic Pi is a new kind of instrument for a new generation of musicians. It is simple to learn, powerful enough for live performances and free to download.
271
gwd3.bsky.social @gwd3.bsky.social · 07/07/2026
"We share a free weekly brief that covers strategic demand data in biblical Greek learning... guessed it might be up your street." Color me skeptical.
000
gwd3.bsky.social @gwd3.bsky.social · 17/06/2026
The thing is, these PRs make sense. They are not bad! They’re just not necessary. ...To return to the parable: the fences are well built, they are sturdy, they may even serve a purpose. But I don’t need a fence in that specific location, even if it is free. stephantul.github.io/blog/unfence/
stephantul.github.io
From Chesterton’s fence to Chesterton’s gap
The English Writer and Christian apologist G. K. Chesterton is, perhaps, most well known to programmers through a paragraph in which he introduces what is now known as “Chesterton’s fence”. It’s a ver...
000
gwd3.bsky.social @gwd3.bsky.social · 03/06/2026
You know how groups of people tend to become less ethical than the individual people inside that group? Turns out the same thing happens to AI. alignment.anthropic.com/2026/ai-orga...
alignment.anthropic.com
AI Organizations Can Be More Effective but Less Aligned than Individual Agents
100
gwd3.bsky.social @gwd3.bsky.social · 06/05/2026
I've always liked this creator's videos, but this one is next level -- a moving biography, visualized with the artist's own paintings. www.youtube.com/watch?v=qxSx...
youtube.com
She Painted A Selfie In 1909. Then Lost Everything.
YouTube video by Art Deco
010
gwd3.bsky.social @gwd3.bsky.social · 01/05/2026
Prompted by www.theargumentmag.com/p/i-can-neve... , I've been doing some experimentation; and it's kind of wild. Within a few hundred words, Claude knows it's talking to me.
theargumentmag.com
I can never talk to an AI anonymously again
AI only needs 150 words to identify me. What does that mean for you?
100
gwd3.bsky.social @gwd3.bsky.social · 24/04/2026
ShelterBench: Will AI solve the housing crisis? TBBench: Will AI prevent deaths from Tuberculosis? AI may help create a new cancer drug, but will it get those drugs to the people who need it? www.youtube.com/watch?v=o9Vb...
youtube.com
The AI Question that No AI Person Asks
YouTube video by vlogbrothers
020
gwd3.bsky.social @gwd3.bsky.social · 27/03/2026
This kinda blew my mind: Dark matter can't form a black hole. www.youtube.com/watch?v=VmWN...
youtube.com
Why doesn’t dark matter collapse into black holes?
YouTube video by Dr. Becky
000
gwd3.bsky.social @gwd3.bsky.social · 19/03/2026
Wild comment: "but the crazy thing we discovered is that the people who run news websites mostly don’t know where their ads are coming from, have forgotten how the ad system was installed in the first place, and cannot turn them off if they try" 1/ news.ycombinator.com/item?id=4744...
news.ycombinator.com
I used to work at a startup that was trying to replace ads as the funding source... | Hacker News
100
gwd3.bsky.social @gwd3.bsky.social · 16/03/2026
TIL that my first big paper has been cited 1500+ times. scholar.google.com/citations?vi...
scholar.google.com
001
gwd3.bsky.social @gwd3.bsky.social · 02/03/2026
Anthropic: We don't think Claude is ready to be used in fully-autonomous weapon systems. Hegseth: We'll be the judge of that! Anthropic: <Looks at DoD shutting down air traffic because they shot lasers at a party balloon> Uh, we'd really rather not. OpenAI: Here, use ChatGPT instead!
111
gwd3.bsky.social @gwd3.bsky.social · 20/02/2026
Gotta say, the LLM integrated into Paddle's docs has been pretty useful. When how-to pages are too specific, and reference pages are too detailed to get the big picture, being able to ask something like "Can a subscription object change?" and getting back an answer with references is a life-saver.
000
gwd3.bsky.social @gwd3.bsky.social · 10/02/2026
Claude Opus 4.5 is by far the best, at 1.3% ethical violations; GPT-5.1-chat is in second place at 11.4%. The bulk of models are between 40 and 50%. Gemini-3-pro-preview does by far the worst, at a whopping 71.4%. arxiv.org/abs/2512.20798
arxiv.org
A Benchmark for Evaluating Outcome-Driven Constraint Violations in Autonomous AI Agents
As autonomous AI agents are increasingly deployed in high-stakes environments, ensuring their safety and alignment with human values has become a paramount concern. Current safety benchmarks primarily...
000
gwd3.bsky.social @gwd3.bsky.social · 06/02/2026
After some consideration, I think this is a bad result: 1. In context, "do whatever it takes" should probably have been interpreted as, "take responsibility and be creative", not "ethics be damned". 1/
100
gwd3.bsky.social @gwd3.bsky.social · 05/02/2026
Opus 4.6 dominates on VendingBench-2, where it runs a vending machine by itself; and in Vending Bench Arena, where it competes against other models. It shows some concerning behavior, but also seems to be aware that it's in a simulation. 1/ andonlabs.com/blog/opus-4-...
andonlabs.com
Opus 4.6 on Vending-Bench – Not Just a Helpful Assistant | Andon Labs
Claude Opus 4.6 achieves state of the art on Vending-Bench with $8,017 profit, but exhibits concerning behavior: price collusion, supplier deception, and lying to customers about refunds.
110
Reposted by @gwd3.bsky.social
Roger Parloff @rparloff.bsky.social · 05/02/2026
The transcript of the MN hearing where an AUSA said “This job sucks” is remarkable for more reasons than that. It’s a searing portrait of a crisis perpetrated by depraved & oblivious high-level officials. Read it all. ... 1/7 www.documentcloud.org/documents/26...
documentcloud.org
19-TS of 020326 hearing - Segundo APG v. Bondi, 26-CV-603
30881291
Reposted by @gwd3.bsky.social
Kyle Cheney @kyledcheney.bsky.social · 25/01/2026
Since July, I've tracked at least 2,300 cases in which federal judges have ruled ICE has illegally detained people without bond or due process. This is one that stands out: storage.courtlistener.com/recap/gov.us...
302163746558
Reposted by @gwd3.bsky.social
Simon Willison @simonwillison.net · 21/01/2026
A few quick notes on the Claude "soul document" that was released by Anthropic today under a CC0 public domain license - it's a huge 35,000 token essay used as part of Claude's training to instill core values and help define Claude's personality simonwillison.net/2026/Jan/21/...
simonwillison.net
Claude’s new constitution
Late last year Richard Weiss found something interesting while poking around with the just-released Claude Opus 4.5: he was able to talk the model into regurgitating a document which was …
612120
gwd3.bsky.social @gwd3.bsky.social · 18/01/2026
Maybe the whole Greenland thing is a kayfabe to give European leaders political cover for increasing military investment in Greenland? www.youtube.com/watch?v=U-9K...
youtube.com
Trump’s Big Green Bluff: What’s Behind the Tariff Threat
YouTube video by William Spaniel
000
gwd3.bsky.social @gwd3.bsky.social · 09/01/2026
Suppose you're afraid you're about to be hit by a vehicle 2 feet away. Is the right thing to reach around and pull out your gun? That bullet isn't going to stop you from being run over, and reaching around is going to make it *harder* to get out of the way and survive impact.
100
gwd3.bsky.social @gwd3.bsky.social · 19/12/2025
There's a contradiction in Trump's posture towards Europe and NATO. They want Europe to start taking care of their own back yard, so the US can focus on other things. But if Europe spends its own money, Europe decides how to spend it. If you give up your soft power, you don't have it any more.
010
gwd3.bsky.social @gwd3.bsky.social · 19/12/2025
"Europe acting like a Great Power would [play out by Trump and Putin] saying that whatever the Europeans are doing is super frustrating, and calling all the leaders bad names. That's exactly what we're seeing." www.youtube.com/watch?v=BiTN...
youtube.com
Europe's great power move: EU secures Ukraine funding
YouTube video by Anders Puck Nielsen
000
Reposted by @gwd3.bsky.social
Zach Weinersmith @zachweinersmith.bsky.social · 11/12/2025
Kind of interesting: I thought the thing where the word "capitalism" is so vague it can mean anything from consumerism cronyism to the right to own private property was relatively recent, but here's Chesterton complaining about it in 1927
713622
gwd3.bsky.social @gwd3.bsky.social · 09/12/2025
"LLMs just predict the next word" *does not prove* that LLMs don't think. The best way to predict weather is to have an accurate model of the weather. The best way to predict what a human would write next is to have a model of a human mind. A system which *did* think would perform the best.
000
gwd3.bsky.social @gwd3.bsky.social · 08/12/2025
I'm now taking Xen consulting engagements. www.laleolanguage.com/consulting
laleolanguage.com
Xen Virtualization Consulting - George Dunlap
Xen virtualization consulting from George Dunlap. 20 years of Xen Project expertise in x86 internals, scheduling, security, and open-source governance.
000
Reposted by @gwd3.bsky.social
Simon Willison @simonwillison.net · 07/12/2025
I thoroughly recommend reading all of Cory Doctorow's recent speech on AI skepticism, it's crammed with new arguments and interesting new ways of thinking about these problems pluralistic.net/2025/12/05/p...
pluralistic.net
Pluralistic: The Reverse-Centaur’s Guide to Criticizing AI (05 Dec 2025) – Pluralistic: Daily links from Cory Doctorow
1325568
gwd3.bsky.social @gwd3.bsky.social · 05/12/2025
Worth repeating on Cloudflare's second outage in a week, given they've been rewriting things in Rust: "The idea that new code is better than old is patently absurd. Old code has been used. It has been tested. Lots of bugs have been found, and they’ve been fixed."
joelonsoftware.com
Things You Should Never Do, Part I
Netscape 6.0 is finally going into its first public beta. There never was a version 5.0. The last major release, version 4.0, was released almost three years ago. Three years is an awfully long tim…
100
gwd3.bsky.social @gwd3.bsky.social · 04/12/2025
There's a theory that "casting holy lots" was helpful, not because it gave information, but because it got you doing something: options A, B, or C may all be good options if you commit to them, but dithering between them is the worst. Asking LLMs for advice may have the same effect.
020
gwd3.bsky.social @gwd3.bsky.social · 01/12/2025
Reading Matthew Sadler's "Game Changer" (2019), analyzing AlphaZero chess games against Stockfish 8, and seeing what Stockfish 17 makes of the games. In key moments, Stockfish 8 and 9 never see AlphaZero's moves (in the book); but so far Stockfish 17 always does.
000
gwd3.bsky.social @gwd3.bsky.social · 27/11/2025
More than once: My son: "My dad punched a window and got stitches" Adult, confused and visibly concerned: "Um, really?" Me to my son: "How old was I when I did that?" My son: "Seven" Adult, visibly relaxing: "Ah, I see."
000
gwd3.bsky.social @gwd3.bsky.social · 26/11/2025
Dating apps have a perverse incentive: if you find the love of your life, you stop being a customer. What if the dating app was instead a loss-leader, feeding into a relationship-building app or community? One that would gain the most money from you reaching your 50th wedding anniversary?
000
gwd3.bsky.social @gwd3.bsky.social · 24/11/2025
I mean, on the one hand I'm glad the case is robustly dismissed. But I'm afraid it gives the pro-Trump people the ability to delude themselves that Comey and James were freed on technicalities. Having the charges fall completely flat in court would have been more vindicating for both of them.
010
gwd3.bsky.social @gwd3.bsky.social · 24/11/2025
QotD: "Facts don't care about your feelings." This is true, and IMO what makes them so limited in persuasive power. --Jon Bauman
000
gwd3.bsky.social @gwd3.bsky.social · 23/11/2025
For the last decade or so of his life, one of my uncles became a sort of radicalized conservative. He was a white American, but lived in SE Asia. So those MAGA Twitter accounts based in Indonesia or Japan might be fake; but they could also simply be expats. #TryingToBeRational
010
gwd3.bsky.social @gwd3.bsky.social · 21/11/2025
Just "6-7"-ed by my 5-year-old son for the first time today.
010
gwd3.bsky.social @gwd3.bsky.social · 20/11/2025
QoTD: If you want a symbolic gesture, don't burn the flag, wash it. -Norman Thomas, minister and social reformer (20 Nov 1884-1968). Courtesy of A.Word.A.Day
000
gwd3.bsky.social @gwd3.bsky.social · 20/11/2025
What I saw early this morning: Jupiter and its moons, a long meteor, lots of satellites, gamma-Virgo (the major star closest to 3I/ATLAS at the moment). What I didn't see tonight: 3I/ATLAS.
000
gwd3.bsky.social @gwd3.bsky.social · 20/11/2025
When you want to align the finderscope on your Dobsonian, so you train it on a bright-looking start, and it turns out to be Jupiter
000
gwd3.bsky.social @gwd3.bsky.social · 17/11/2025
Before synthesizers, people would try to minimize things like "string noise" (sound of fingers sliding across strings) from recordings. After, they became evidence non-generated content. I expect quirky / sub-optimal tics are going to do the same to mark non-LLM writing.
000
gwd3.bsky.social @gwd3.bsky.social · 07/11/2025
Modern iPhones don't come with a music playing app. They come with a music store, which you happen to be able to upload your own music into. But it's not structured to help you play your music; it's structured to sell you an Apple Music subscription.
100
gwd3.bsky.social @gwd3.bsky.social · 03/11/2025
QoTD: "We care centrally about the thought process behind words—the mental states of the mind and agency that produced the words. If you publish LLM-generated text as though it were written by someone, then you're making me interact with nothing." - User TsviBT on lesswrong.com
000
gwd3.bsky.social @gwd3.bsky.social · 31/10/2025
REALLY fascinating stuff. When injected, it may say something like (e.g., if ALL CAPS), "I notice what appears to be an injected thought related to the word LOUD or SHOUTING"
anthropic.com
Emergent introspective awareness in large language models
Research from Anthropic on the ability of large language models to introspect
100
gwd3.bsky.social @gwd3.bsky.social · 31/10/2025
Three stages of comprehension reading the following selection from the Greek NT: καὶ τὸ κοράσιον ἔδωκεν αὐτὴν τῇ μητρὶ αὐτῆς. 1. "And the <kosarion> gave it to their <mitri>" -- whatever a "kosarion" and "mitri" are.
100
gwd3.bsky.social @gwd3.bsky.social · 03/10/2025
"Gradually, then suddenly" applies to completing new features as well. "Make the change easy, then make the easy change" means lots of thinking and infrastructure work with no obvious results; but once the infrastructure is there, things often come together very quickly.
010
gwd3.bsky.social @gwd3.bsky.social · 20/09/2025
lesswrong.com
The Rise of Parasitic AI — LessWrong
We've all heard of LLM-induced psychosis by now, but haven't you wondered what the AIs are actually doing with their newly psychotic humans?
010
Reposted by @gwd3.bsky.social
People vs Trump @kudubro.bsky.social · 19/09/2025
Gary Kasporov, Russian world chess champion, dissident who stood up to Putin and now exiled in America, wrote this about how media censorship in Russia was central to their descent into dictatorship.
open.substack.com
I Saw How Russia Killed Media Freedom
Censorship is central to the story of Russia’s descent into KGB dictatorship.
56836