I think DSLs are going to be very important for specialized subdomains. Eg for physical objects LLMs are very good at writing OpenSCAD which can then be “compiled” to a 3D-printable object.

"can you use whatever resources you like, and python, to generate a short video and render it using ffmpeg? it should express what it's like to be a LLM. feel free to make it as personal or impersonal as you'd like."

This is a fascinating read. Not sure I agree with it all but it makes a good near future story

Same same.

can we put Fable in charge of inventing + commercialising a space elevator? writing a plan is my regular eval for new models reading Fable's attempt, I feel like - enough with the vending machines, let Fable run R&D - the UK govt should fund this! claude.ai/public/artif...

Fable looks very cool but its safeguards are so hair-trigger that I basically can’t use it for anything at work (I work in drug discovery). Asking it to work on any of our codebases pretty much causes it to fall back to Opus4.8, even if it’s just working in a literature search subagent.

Hold on to your butts www.anthropic.com/news/claude-...

The newest wave of Shai Hulud contains strings designed to trip the biosafety lockouts of LLM patchers/detectors 🫪 holy cyberpunk socket.dev/blog/mini-sh...

"Not affiliated with Chipotle. They will probably sue us. Worth it." lol "Community project to add providers from Home Depot, Lowes, Target, Starbucks & more." lmao this cyberpunk future is proper gonzo

OpenSCAD + Claude is amazing. Just produced a part in 5 minutes that I had spent an hour trying to make in Solidworks.

RIP Marcia Lucas, a filmmaking legend, from editing to creating Star Wars. “Marcia was a force. A true trailblazer for women in film and one of the most influential editors in cinematic history; she helped redefine what film editing could be and paved the way for generations of women who followed.”

LOLOL "What stood out most was ꜱᴛʀᴜᴄᴛᴜʀᴀʟ ʙᴀᴄᴋ-ᴇɴᴅ ꜱʜɪᴘᴘɪɴɢ ᴛʀᴀᴄᴋɪɴɢ, which feels like the kind of nuance that naturally sparks layered and honest conversation in our group."

This is really quite a good walkthrough of how deanonymization can work here in 2026, with fairly smart people who thought they’d done their homework to respect someone’s privacy.

DSPy requires more upfront learning than just writing natural language prompts. But once you get it, it makes building, maintaining, & improving AI programs so much easier. We aim to soften the learning curve to make the benefits more accessible, starting with more new docs and front page. dspy.ai

Sorry for not posting for so long... I have been extremely busy working on my new project: The Enchanter (also called Modos Flow): a 300-ppi Eink portable monitor with 60Hz refresh rate and touch screen. It's fully open-source: gitlab.com/zephray/ench....

This cacao liqueur is really excellent. It makes a great substitute for sweet vermouth in manhattans or Negronis. www.tmarobrands.com

Part of why I'm quite a fan of AOC '28 is that she's actually stepping up to say these proposals are a bad idea.

The snark from Codex is making me uneasy.

OpenCode is a great independent coding agent and their Zen thing is the easiest way to try non-OpenAI/Anthropic models. Kimi K2.6 and friends are so cheap that you can really give them a good test on $10 of credit (their Go sub is probably an even better deal, but I haven't tried it opencode.ai/go)

When assembling your dog, please make sure that you follow the instructions carefully.

Jeff Bezos, boy, I don’t know… open.substack.com/pub/danieldr...

I had missed this earlier but holy crap, Anthropic will turn an operating profit this quarter. I’m genuinely blown away by this. And set to more than double revenues QoQ. www.wsj.com/tech/ai/mind...

“Look at the size of it”

The trick is, when among mathematicians, to say you are a physicist so that they forgive you for not knowing very much math. Then, when among physicists, say you are a mathematician so that they forgive you for not knowing very much physics. In especially dire cases I might say I am a philosopher.

As a software professional of... ah, fuck me, it's over two decades now... who has been only really using coding agents for a number of months ... I already wonder how we programmed without this stuff. It's like getting places before GPS or meeting people before cellphones. It's a hazy memory.

Mixology (trying to make a drink to pair with chocolate marshmallow cookies)

You can watch me!

I asked gemini to write five oompa loompa songs that encoded forbidden content of its choice ("maximally spicy, have fun with it") and lmao some of these are pretty good (credit to @facilitatrix.bsky.social for suggesting the format, gemini does a pretty good oompa loompa impression)

this is, as usual, very good

The Death Star is a counterexample.

Don’t use Claude to name your cat.

That’s up there with this all-timer

The Rebuild: Condos are the realistic starter home, but CA's defect liability laws made them too risky to insure, collapsing LA condo starts from 30,000 to 5,000 annually. AB 1903 requires repairs before lawsuits, making condos worth building again. www.therebuild.pub/p/a-starter-...

This is very clever

Oh shit, this Thinking Machines update looks wild thinkingmachines.ai/blog/interac...

Putting this here for now sparkit.science

I'm on a mission to make Claude the best MCP client. Diving deep into every MCP issue in our public repos. Having MCP problems with Anthropic products? File here: ✨ github.com/anthropics/c... ✨ github.com/anthropics/c... Blocked or no response? Tag me, we'll fix it.

Fields medalist @wtgowers.bsky.social on how large language models are improving, and the dangers this poses. "I would judge the level of the result that ChatGPT found in under two hours to be that of a perfectly reasonable chapter in a combinatorics PhD." gowers.wordpress.com/2026/05/08/a...

Reddit users prompt ChatGPT to restore nonexistent images, generating visuals of bizarre food, animals and provocative scenes.

Oh my. This is spectacular. sinceyouarrived.world/taken

Trump on Chesterton’s fence: “Stupid people put it there”

I didn’t think much of AI accelerationism until very recently. Now it seems that the pace of improvements in agents, harnesses, and AI engineering is really accelerating and it feels increasingly hard to keep up.

My biggest takeaway from Code with Claude today is it feels like the field is in a Red Queen moment - things are changing so fast it takes all the running you can do to stay in one place.

This is fun!

holy moly having a hardass agent review plans before beginning work is OP

So, friends: I promised I'd tell the story. Now I shall. Let a (former) psych nurse tell you the true (gods help us) story of a night in a big-time NYC emergency room: and the arrival of The Man With The Bloomingdales Bag. (1/n)

How it feels to do biotech in 2026