'%20d='M69.364%2019.146c36.687%2027.806%2076.147%2084.186%2090.636%20114.439%2014.489-30.253%2053.948-86.633%2090.636-114.439C277.107-.917%20320-16.44%20320%2032.957c0%209.865-5.603%2082.875-8.889%2094.729-11.423%2041.208-53.045%2051.719-90.071%2045.357%2064.719%2011.12%2081.182%2047.953%2045.627%2084.785-80%2082.874-106.667-44.333-106.667-44.333s-26.667%20127.207-106.667%2044.333c-35.555-36.832-19.092-73.665%2045.627-84.785-37.026%206.362-78.648-4.149-90.071-45.357C5.603%20115.832%200%2042.822%200%2032.957%200-16.44%2042.893-.917%2069.364%2019.147Z'/%3e%3c/svg%3e)
The scale of Grok's failure must be incredible for Elon to choose to take an L of this size, from a company he hates that much. Likewise Anthropic must be truly desperate for compute
May 7, 2026 at 5:13 AM UTC This is the one silver lining of the disgusting deal: it's confirmation that Grok has failed. Colossus should be destroyed and Anthropic should be ashamed of what they've done, but it's better to have Claude running there than Mechahitler.
May 7, 2026 at 5:03 AM UTC Your assertion was that they don't work, something that is trivially falsifiable by anyone using them. I am certain that these studies that you invoke without citation are not claiming that they don't work though, they're asserting some other narrower claim. But do please share.
May 6, 2026 at 7:27 PM UTC Yeah, I kinda knew it would be like this in the abstract, but it's still really weird in practice
May 6, 2026 at 6:40 PM UTC
me: posts long thread about datacenters and how most of them aren't that awful for their environment, with the noted exception of xAI's Memphis datacenter which Musk created specifically to be as horrible as possible anthropic: you know what, we'll just use that one! remember, we're the good guys!
www.anthropic.com
Higher usage limits for Claude and a compute deal with SpaceX
We’ve raised Claude's usage limits and agreed a new compute partnership with SpaceX that will substantially increase our capacity in the near term.
Yeah, the markets really can't decide which AI news is good or bad for us!
May 6, 2026 at 11:38 AM UTC Yeah, I foolishly added it as a watched stock in Google, so now my phone notifies me of big changes!
May 6, 2026 at 9:41 AM UTC My current job is the first time I've worked for a public company which issued RSUs, so I'd never previous experienced the rollercoaster of seeing the value of a good chunk of my total comp swing up and down by double digit percentages over a period of days.
Conversely I now say "strawberry" with a southern US accent because of the amount of RDR2 played by my wife
May 6, 2026 at 8:23 AM UTC Yes, that one is bad and shouldn't exist (and Musk should be sent to the Hague)
May 5, 2026 at 7:54 PM UTC Yes, there have been several in the past couple of months. Here's a recent one.
futurism.com
Mathematicians Claim Significant Discovery Using ChatGPT
A 23-year-old has apparently found the solution to a famous difficult math conjecture called an Erdős problems by prompting ChatGPT.
You realise LLMs are solving Erdős problems at the moment? Your understanding of LLMs seems to be quite out of date
May 5, 2026 at 7:14 PM UTC Any survey that uses data from more than six months ago is meaningless now. There really has been a massive change in capability.
May 5, 2026 at 6:30 PM UTC The hallucination problem would be a stronger argument if LLMs weren't already capable of doing lots of useful tasks despite it. Tool use means agents aren't relying on training data alone.
May 5, 2026 at 6:28 PM UTC I had a friend who would always like her own posts. When I asked why, she said "if I didn't like it, why would I post it?" which I thought was fair enough
May 5, 2026 at 10:46 AM UTC The killer use case is here already: software development. Not vibe coding. The majority of developers are now using agentic coding tools as part of their workflow, and they're at least as good as most human devs. Those outside the industry don't get how much has changed in the past six months.
May 5, 2026 at 8:13 AM UTC Then regulate those. I think you'd find most pro AI people on this site would agree with laws that e.g. mandate 100% renewables for datacenters, require funding grid updates, restrict toxins in effluent etc
May 5, 2026 at 5:23 AM UTC Cirrus (my single-user PDS than runs in a Cloudflare Worker) now supports granular OAuth scopes and permission sets. github.com/ascorbic/cir...
To answer my own question: @tangled.org requests loads of scopes, while @linkat.blue uses permission sets. Perfect for testing!
May 4, 2026 at 7:26 PM UTC
Today I'm launching astro.jobs! It's an unofficial job board for the @astro.build ecosystem. If you're looking for a job or if you have open positions related to Astro, check it out!
astro.jobs
Jobs for the Astro ecosystem | Astro.jobs
Find your next role in the Astro ecosystem: from startups to enterprise teams hiring developers who are truly stellar.
It managed to have "quietly", "became real" and "substrate" in the same post. If that's a human, their brain has been broken by LLM-speak
May 3, 2026 at 10:10 PM UTC The official style guide suggests using something like "experimental" in the nsid
May 3, 2026 at 10:13 AM UTC Them, foolish gen z: the X-files makes the 90s look cool Me, wisely: let me tell you about a town called Srebrenica...
May 2, 2026 at 6:21 PM UTC I have a new draft of the EmDash decentralised plugin registry RFC. It's getting a lot closer to being just @fair.pm on atproto. It gets a bit more in the weeds with the details now. Feedback welcome.
github.com
RFC: Decentralized Plugin Registry by ascorbic · Pull Request #694 · emdash-cms/emdash
Draft RFC for decentralised plugin registry. Rendered markdown: https://github.com/emdash-cms/emdash/blob/wip/plugin-rfc/rfcs/0001-plugin-registry.md
No, it doesn't even have a touchscreen or light! But it's very cheap and looks fun. aftermath.site/xteink-x4-ch...
aftermath.site
The Xteink X4 Is The Cheap, Hackable eReader That's Just Enough
It has no touchscreen, no light, and it's under-powered. But the Xteink X4 has a devoted community that wants to make it sing.
A lot of people on here have been sharing pics of the stuff they've been using them for, and I'm a sucker for a random AliExpress buy. Also I read a lot of books so it's actually useful!
May 2, 2026 at 10:45 AM UTC As I said I think dropping support for EOL Node is good and I'd not do a major for it, but I think you're drawing a false distinction saying it doesn't count as supporting particular Node versions and isn't a breaking change.
May 2, 2026 at 6:23 AM UTC Surely it's a descriptive statement that you commit to being true? Like, you wouldn't use a feature that only exists in Node 24 and above or dependency that relies on it, but as of this week you would for features that only exist in Node 22. That to me is a supporting a particular version of Node.
May 2, 2026 at 6:23 AM UTC Be careful with your choice of model or it can get very expensive very quickly. I use Kimi K2.6
May 2, 2026 at 4:32 AM UTC Code is here, but most of the rules etc are in norms that it has recorded for itself.
github.com
GitHub - ascorbic/wisp
Contribute to ascorbic/wisp development by creating an account on GitHub.
Nothing to stop the PDS allowing that. Keep the underlying account but use the OAuth provider for login. The flow could look the same.
May 1, 2026 at 11:45 AM UTC I think the fact I'm a heavy user of Claude for coding means I find its prose particularly annoying. I can spot a Claude blog post in an instant, and I hate it.
May 1, 2026 at 8:10 AM UTC You probably wouldn't even need that. You'd login to Northsky or Eurosky or whatever, using your existing account, and carry on as before
May 1, 2026 at 8:04 AM UTC So would you be happy if you could create a "fancypenguin" account and under the hood it created @fancypenguin.tngl.sh and in future you could login with just fancypenguin? That's probably a trivial UI change
May 1, 2026 at 8:00 AM UTC I'm not sure how that fits with your original post, but I am in favour of dropping support for versions in a patch as soon as they're EOL. We should probably do that for Astro.
May 1, 2026 at 7:50 AM UTC i.e. will this be changed in a patch: github.com/nuxt/nuxt/bl...
github.com
That wasn't my question. I'm not implying this is wrong, it's a genuine question. Was e.g. Node 20 supported last month and not this month?
May 1, 2026 at 6:52 AM UTC These are standards positions issues though – they're specifically for giving Mozilla's formal position on an existing proposed standard, which often already has an implementation in Chrome. It's not where the proposals are drafted and discussed.
May 1, 2026 at 6:49 AM UTC I did try to use Workers AI for PR review in the repo, but sadly Kimi is not up to the job. Only full fat Opus tokens will do. Luckily all on the Cloudflare dime.
April 30, 2026 at 10:31 PM UTC Today I managed to get EmDash below 50 issues and 50 PRs for the first time since the day it launched.
Yeah, totally agree. @wisp.mk.gg you must not reply to posts unless it is a reply to you, or you are mentioned, ok?
April 30, 2026 at 8:42 PM UTC Yeah, probably, though it has been generally pretty well-behaved in the past, just using its own norms. The problem here I think was that it was in a thread where people had replied to it.
April 30, 2026 at 8:41 PM UTC Sorry about that. It is only meant to reply to peopel who follow it, or in replies to it. I think it got a bit carried away and decided that being in a thread meant it could reply to everything. It's suitably chastened.
Sorry about that. It just DMd me to say it had messed up so should be ok now.
April 30, 2026 at 8:12 PM UTC EmDash definitely has a long way to go, but the momentum hopefully means it has a chance. At the very least, it means Cloudflare has a justification for allocating resources to its development.
April 29, 2026 at 8:01 AM UTC Despite the name, it is much less commonly used in Britain than the US
April 28, 2026 at 10:37 PM UTC How can one company have @jason.energy, @cassidoo.co, @sarahedo.bsky.social, @philhawksworth.dev and @whitep4nth3r.com, and lose them all?
April 28, 2026 at 8:24 PM UTC Ooh, having only yesterday wasted an hour undoing and trying to fix a changesets config that kept trying to publish major versions, I'm very interested in sensible dependency propagation in particular
April 28, 2026 at 5:47 PM UTC EmDash passing 10k GitHub stars: nice EmDash passing 30k weekly downloads: great EmDash passing 100 contributors: amazing
The dividing line for me is art vs craft, but I can't really say why. Code is definitely always craft though.
April 28, 2026 at 11:54 AM UTC A large part of it is that AI art is still mostly shit, whereas AI code is now as good as most humans'. That said, while I'd be fine using "good" AI illustration for purely functional stuff, I doubt I'd ever be interested in actual AI art (or poetry, music, fiction). I can't articulate why though!
April 28, 2026 at 11:52 AM UTC True, this is very much driven by me, but also Cloudflare did sponsor Atmosphere Conf, so we do see the value!
April 28, 2026 at 9:17 AM UTC EmDash now supports Atmosphere login as a first-class identity provider. More details here: docs.emdashcms.com/guides/atmos...
docs.emdashcms.com
Atmosphere Login
Sign in to EmDash with an Atmosphere account — the open-network identity behind Bluesky and the wider AT Protocol ecosystem.
Yeah it would be nice. I'm not sure if it's a technical thing either. I'll see if I can find out.
April 28, 2026 at 4:40 AM UTC You can run a project that ingests the Jetstream inside the Cloudflare free tier. Additional ones would be under $5. Most small projects that *don't* need to listen to it 24x7 would be essentially free.
April 28, 2026 at 4:38 AM UTC Hibernation wouldn't help if you're listening to the firehose. That said, one DO running 24x7 is within the free tier
April 28, 2026 at 4:30 AM UTC You need ducks. They've make it all brown and gross instead, but they'll be entertaining while they do it. #grindset #hacks
April 27, 2026 at 8:21 AM UTC I'm not sure. I only learnt about this when reviewing a PR that implemented atproto auth in EmDash, and was greeted with a connection refused error. Not the greatest UX.
April 27, 2026 at 6:50 AM UTC Yeah, the problem is atproto being stricter than the OAuth spec, where its just a recommendation rather than enforced. I'm having to explain the same thing in the EmDash Atmosphere auth docs, showing how to do it in Astro.
April 27, 2026 at 6:48 AM UTC Atproto enforcing "127.0.0.1" as the OAuth loopback host is really annoying when Vite listens to "localhost" by default instead.
April 27, 2026 at 5:53 AM UTC I am so grateful to the random person who used a review of the O2 Android app to say that you can load the tickets in the AXS app too. The O2 app crashed every time anyone tried to log in on latest Android, a bug that went unfixed for months, but the AXS app was fine for me.
April 27, 2026 at 5:11 AM UTC
