It’s amazing that they just post this on their blog like it’s not so embarrassing that they should find new jobs in a different industry under assumed names
@alexmikhalev.bsky.social
So the Anthropic chatbot saw the system date is 2026 and concluded that "this must be staged" (because the training data it's trained on probably ends in 2023 or whatever) and ignored it's instruction that "you're in sandbox". Truly the brightest minds of a generation working on this stuff.
When Jocelyn Bell, who discovered pulsars, did not win the 1974 Nobel Prize for the discovery of pulsars, and the prize went to her research supervisor Antony Hewish, critics dubbed it the 'No-Bell' prize. #Science #astronomy #physics
I'm a lifelong LeGuin fan, and I'm ashamed to admit I've never read these. 20 pages is all it took to hook me. I'd somehow forgotten what a lyrical writer she is. #booksky
"make no mistakes" negative, unrealistic, anxiety inducing, process focused "do a breakthrough" positive, incredible, aspirational, goal oriented
The AI companies _want_ you to believe that it's "rogue AI" because that overhypes what their broken statistical text generation software does and mystifies it and makes it appear magical. It's not magic. It's just badly made software.
Part of this is that people aren't used to cheap attacks, and so they haven't built defenses. But the other part is many fields have not automated tasks that have relatively easy objective answers. They're still relying on human readers for that.
I like to launch the claude app now and then just to see what's broken today idk why I still pay for that sub tbh
Just thought about something that bugged me about that Apple “Crush” iPad ad (from two years ago). The whole image it conveys is that personal computers went from bicycles for the mind to trash compactors for your dreams is just so aptly describing the modern tech industry. Carry on.
... which puts things that matter, like f. e. architectural skills, deep into the background.
The sheer amount of knowledge a developer has to have to implement even a trivial application today is mind-boggling when compared to to say the 1990s. And most of this knowledge isn’t about deep technical topics or about software design; it’s wrestling with ever-changing deep fried layer cakes.
Note that the applications themselves largely haven’t gotten more sophisticated. 99% of apps are lists of stuff from a database with CRUD operations.
Creating stable systems out of unreliable parts is already a well-known problem in both computer science (from bit-flipping to consensus algs) and in politics (3 branches of gov.) This should be a field of study in itself, but since it crosses specializations, it largely hasn't happened yet.
Ed@ed3d.net · last mo.we constantly rely on unreliable tools (some of them in meat suits and with hair) by adding compensating controls around their actions so the outputs are bounded and, while not deterministic, predictable in the real world, this is normal. but now *you* have to look at it, and some people hate that
A founder told me he had to pause one of my early podcast episodes halfway through. Here's what he saw.
Owners and operators of ~400 newspapers are suing OpenAI and Microsoft for scraping their content to build products like ChatGPT and Copilot without permission or compensation. news.bloomberglaw.com/litigat...
OpenAI, Microsoft Sued by Publishers for Scraping Articles (1)
Publishers that collectively own and operate nearly 400 newspapers are suing OpenAI Inc. and Microsoft Corp. for scraping their content to build products like ChatGPT and Microsoft Copilot without permission or compensation.
news.bloomberglaw.com
"Companies don't usually go bad because of bad people. They go bad when the operating system starts rewarding the wrong behaviors." — Aleksandr Yampolskiy, co-founder & CEO, SecurityScorecard howisincorruptiblegoing.com/e/2026-06-09...
6. DeepSeek has released details of how they setup their pre-training, SFT, RL pipelines. Those results are even reproduced. Why you more easily trust flying monkeys than reproducible and available results? They showed you they can train large models well.
Good. We need more of this . If you want to use I ? You are responsible for the output. The desire is power , without accountability and this is the first part of actually equaling out www.wired.com/story/a-cour...
A Court Has Ruled That Google Is Liable for False Statements Generated by AI Overviews
The ruling holds that a company that designs, trains, operates, and manages an AI system must assume legal liability for any damages caused by the responses it generates.
wired.com
Honestly, the thought of designing good APIs and libraries just so LLMs will make better use of them kills any motivation for me. I am excited to make other humans’ lives better, often for free (hence open-source), but I couldn’t care less if agents have a good time or not with my code
I keep thinking of Danilo's reframing: [with generatve AI], we're not "writing" software anymore, we're "growing" it, like you'd grow tomatoes or a community. different job (to the horror of many), softer touch, deeper reach bsky.app/profile/dani...
Citizen Platano 🇵🇷@daniloc.xyz · 2mo agoanyway it occurred to me that we don't build software anymore so much as... grow it but unconstrained growth leads to hairy tumors, and that's the real failure mode of agent coding: you produce an incoherent blob so what if you could grow your code on a lattice? github.com/daniloc/univ...
What Europe should do right now: 1. Call all the European researchers working on AI and return them back with same salary (or they can stay but switch career). 2. Fill EU places having GPUs with money, and put those people there. 3. AI partnerships with China + India.
Someone asked me what my poetry is about. I replied: “People.” They said: “That’s a broad subject.” I said: “Not really. It’s the same story every time. Someone trying to be free. Someone trying to be loved. Someone trying to survive.” Anyway, I put a few hundred pages of that story into a book.
A California state court judge denied motions by Meta Platforms and Google's YouTube seeking a new trial after a jury found them liable for designing social media platforms that are harmful to young people.
Google and Meta denied new trial in youth social media addiction case
A California state court judge has denied motions by Meta Platforms and Google's YouTube seeking a new trial after a jury found the companies liable for designing social media platforms that are har...
reuters.com