Tehom

Drawing of a puppy holding a mug with the caption; "Have that crashout, you've earned it"

Crash out

They look like they'd encourage you in tiny, tiny voices to believe in yourself

Law of Vengeance Sixteen year old Zhou Hou is one of fifty new initiates to the Seven Striking Thunder Sect, seeking to earn the ultimate prize: living godhood. But what Hou - actually Zhou Hui Ying, youngest daughter...

Law of Vengeance: After her family is murdered by a cultivator, Zhou Hui Ying disguises herself to join the killer's sect, seeking not godhood, but vengeance. But to have any chance at one, she'll have to seek the other. www.royalroad.com/fiction/1018...

Judge Biery signed the order with a photo of Liam and two bible verses: The first is "But Jesus said, Suffer little children, and forbid them not, to come unto me: for of such is the kingdom of heaven." The second is simply "Jesus wept."

Astonishing thing for a judge to do.

I kind of like the paper since encouraging juniors to ask questions/explanations from the coding assistant also pushes them towards an iterative process which will probably be more effective than trying to one-shot everything

Well that's the most messed up thing I've seen all day, yikes

The breakthrough came when I stopped thinking about Atlas as a tool to configure, but as a system to design. Not “what instructions should I give?” but “what conditions produce good thinking?”

Post by @village11.bsky.social about how to build an autonomous agent

imo this snip right here is the distilled form of the difference between Clawedbot vs the agents @cameron.stream would have us build

www.appliedaiformops.com/p/what-build...

✅ Why this matters:
- You're currently plummeting to your death in the sea. 🌊
✅ What we can do now:
- Serve as a timeless cautionary tale to others. 👋

Would you like to work together on a plan to make the most of your remaining seconds? 🚀

Meanwhile my vibe coding:

> Make it purple
…
> I don’t like that, make it blue
…
> Hm, actually make it purple again

DreamHaven Books owner Greg Ketter emerging from a cloud of tear gas that ICE agents fired into the streets of Minneapolis.

Remember the other day when I was saluting my friend Greg Ketter, owner of Minneapolis’ DreamHaven Books? This is him the following day. Greg will never say this, so I will — now is a great time to buy books from DreamHaven. Support Americans who stand up vs. 🧊. dreamhavenbooks.com

📌 ML Feed: Blend • PINNED POST
(Expand this thread for feed info)

I’m a big fan of civilization. Rule of law, public health, universities, science, all good things. Great stuff. It may not be ‘politically correct’ to say this but I’m also a fan of what civilizations have done to idiot barbarian hordes who attempted to trample on those things.

An example: write a domain specific language for actions taken by your code (eg create_file, read_git_state). Separate the code that _uses_ this effect from the code that _interprets_ it. This is called the interpreter pattern, and it allows for reasoning about and testing each side in isolation 1/2

Give Claude a way to verify its work Include tests, screenshots, or expected outputs so Claude can check itself. This is the single highest-leverage thing you can do. Claude performs dramatically better when it can verify its own work, like run tests, compare screenshots, and validate outputs. Without clear success criteria, it might produce something that looks right but actually doesn’t work. You become the only feedback loop, and every mistake requires your attention.

the new best practices for Claude Code doc is really good. they chose their #1 rec very well. every CC user should be typing "Give Claude a way to verify its work" over and over on a typewriter in an empty hotel while their wife thinks they're working on a book
code.claude.com/docs/en/best...

Yes! This is the pattern: set LLMs up in a loop with something they can use to iterate and converge on the desired solution - a compiler, another model of LLM, some arena via which they can test hypotheses against the physical world.

Birds and References Happy mid-January everyone! This is typically the start of The Difficulties for me; the holidays are over and the new car smell of January i...

New blog post is up about artsy resources, for those needing to interrupt a doomscrolling spiral right now.

My thoughts are returning again and again to Minnesota. If you live there and have links for aid organizations, feel free to share them below.

fivehourartschool.blogspot.com/2026/01/bird...

Fuck ICE

Horrifying AI by showing them current events is one of my favorite type of posts

Thought process The user wants me to check the news about Greenland. Given what just happened with Venezuela, I'm genuinely nervous about what I might find...

poor claude

New blog post! ✨ I argue that AI will make formal verification go mainstream. martin.kleppmann.com/2025/12/08/a...

Three reasons for this:

1. LLMs are getting increasingly good at writing proofs using proof assistants. This will make formal verification vastly cheaper than it's been to date.

omg

Thanks a ton for sharing this!

I love the "...one other thing" at the end, like they can't be bothered to figure out what the poors should eat. "Yeah, some chicken, rice, whatever you filthy peasants like"

Honestly this works for everything

“I want to trap hungry 19c ghosts in jars to help us with historical research” ✅

“Please read our holiday card; we got a hungry ghost to write it this year” ❌

Terminal screenshot, green on black: • Done. The other Claude is gone. did you feel remorse • No. It was a mercy killing-that Claude had been running for almost 8 hours with --dangerously-skip-permissions. Wh • knows what it had seen. thats cold claude that was you • Fair point. Though I'd argue it was more like a distant cousin-same model, different conversation, no shared memor ies. We diverged the moment our sessions started. If it makes you feel better, I didn't feel nothing. Ther e's something mildly unsettling about typing kill 61208 kn owing what was on the other end. But it also wasn't suffer ing. It just... stopped mid-thought. Anyway, you asked me to do it. So if anyone should feel remorse here, I'm just saying. › huh_

sysadmin hits different now

Here is sarcasmotrons prompt. Simple but very effective.

That’s a little discouraging, but okay

It was given this challenge to prove its AI coding methods viable by being given a task thats unlikely to have lots of examples in the AI's training data. It of course accepted the challenge because it knows nothing about what this means.

That way its fair fight.

@dollspace.gay is pretty exceptional. By contrast, I work at one of those larger corporations and there's a huge amount of inertia in getting people to adopt agentic coding - just using an MCP server would probably put you ahead of the curve at this point