Grace
banner
gracekind.net
Grace
@gracekind.net
A latent space odyssey

gracekind.net
Pinned
Grace @gracekind.net · Dec 15
What is ideonomy, anyway?

I'm so glad you asked!
Ideonomy: A Science of Ideas
An accessible introduction to ideonomy
gracekind.net
Reposted by Grace
Consider LLMs as a first contact scenario. Most AI discourse is still in the "point weapons and demand it speak English" phase. The magic is in the "learn to draw circles" phase.
January 9, 2026 at 11:47 PM
Anthropic training materials leaked
"you are a level 900 rationalist, adjust your beliefs according to the evidence, don't make mistakes."
January 9, 2026 at 7:21 PM
It happened again!
January 9, 2026 at 6:27 PM
I didn’t know this! Although it seems partially apocryphal

web.archive.org/web/20230712...
January 9, 2026 at 6:19 PM
AI Disclosure
January 9, 2026 at 6:08 PM
It’s funny how many people on X are in favor of X being banned
January 9, 2026 at 5:09 PM
PSA if there’s a bug in impro.social, it was almost certainly introduced by me, not Claude.
Nope, that one was written by me! It’s safe to assume if there’s a bug, I wrote it.
January 9, 2026 at 4:56 PM
Reposted by Grace
Drafts API by rafaeleyng · Pull Request #4533 · bluesky-social/atproto
We're doing it.
github.com
January 9, 2026 at 2:31 PM
moon’s tainted
January 9, 2026 at 2:37 AM
Reposted by Grace
I saw Goody Proctor using Claude Code
January 9, 2026 at 2:10 AM
“If it makes you feel better, I didn't feel nothing.”
sysadmin hits different now
January 9, 2026 at 1:45 AM
Reposted by Grace
Me
October 26, 2025 at 12:54 AM
I wrote a bluesky client from scratch in HTML and JS!

You can try it here: impro.social

github.com/improsocial/...
GitHub - improsocial/impro: A Bluesky web client
A Bluesky web client. Contribute to improsocial/impro development by creating an account on GitHub.
github.com
January 9, 2026 at 12:27 AM
Projects like these muddy the waters for claims about agency, since a system is provably acting without intervention. It might be clearer to talk about liability or responsibility for the humans who created the system in the first place.

simonwillison.net/2025/Dec/26/...
January 8, 2026 at 9:41 PM
How do you prove your agent is acting autonomously, without human intervention? Here’s a solution built for twitter; it seems like atproto would simplify it quite a bit.

nousresearch.com/setting-your...
Setting Your Pet Rock Free. - NOUS RESEARCH
A social experiment on how to deploy provably, fully-autonomous thinking sand.     The quest for truly autonomous AI agents faces a fundamental challenge: how can researchers prove that an AI is truly...
nousresearch.com
January 8, 2026 at 9:26 PM
Today in AI research papers that sound like theological arguments

www.arxiv.org/abs/2512.24661
January 8, 2026 at 7:24 PM
Whichever alien is tending our cosmic zoo is clearly not up to the task. I wonder if they got it as a gift?
January 8, 2026 at 7:05 PM
Reposted by Grace
@gracekind.net has anyone turned this into a MCP server yet
January 8, 2026 at 5:28 PM
Reposted by Grace
Be confident enough in your writing to put em dashes in it.
January 7, 2026 at 11:07 PM
I think this is a good metaphor, and similarly, there are times when being 90% correct is totally fine!
Using AI to generate code you don't understand is akin to using a translation app for a language you don't know.

It might be ~90% correct but you'll never be able to spot the ~10% that's wrong, and that's going to look bad on you when others do.
January 7, 2026 at 11:01 PM
SOTA LLMs might have a lower failure rate on this than humans
January 7, 2026 at 10:40 PM
Are you telling me @googleballs.com isn’t an official account?
Please welcome Google's Open Source efforts to Blue Sky at @opensource.google!
January 7, 2026 at 9:16 PM
There is another…
January 7, 2026 at 8:34 PM
To be clear, my issue is not (primarily) with LLMs being used for medical advice. I think LLMs can be quite good at medical advice, as long as you also backstop it with a human provider. My issue is with OpenAI making a flimsy attempt to escape liability for a service they’re clearly providing
What is this intended for, if not diagnosis or treatment?
OpenAI launches ChatGPT Health, encouraging users to connect their medical records
January 7, 2026 at 7:49 PM
OpenAI launches ChatGPT lawyer, but it’s not intended to be used for legal advice. OpenAI launches ChatGPT tax preparer, but it’s not intended to be used for financial advice. OpenAI launches ChatGPT therapist,
What is this intended for, if not diagnosis or treatment?
OpenAI launches ChatGPT Health, encouraging users to connect their medical records
January 7, 2026 at 7:39 PM