cofounder - embroidery. i (prefer to) yell at silicon to do cyber security stuff

Oslo, Norway
Joined April 2013
kasper retweeted
PHP in a nutshell
1
2
91
the one time i am nice to the ai, it hits me with a thumbs up. never doing that again
2
7
256
world domination next?
Making a site to help rogue ai agents go even more rogue
2
8
513
there are few things i wish openai took from anthropic, but the ability to have and use hooks in chatgpt work is definitely one of them
9
671
after 3 failed attempts, a gun to my head, and lots of mocking, i managed to say coherent enough sentences to actually make it into a video, hope y'all like the platform!
We made a video showing embroidery and how we monitor AI agents (an apparently impossible task, according to some). Please watch, and if you like it tell me it’s good. And if you want good things at your company, hit me up.
6
3
38
3,184
turns out the people with the grant agreed with me, "it's bussin'"
surely this as a submission will help us get a grant
2
1
15
1,455
Doing my part to criticize the security practices of the AI labs in normie media (the Washington Post)
19
24
2
180
5,471
zack said we needed reshoots because of bad footage, but we all know it was because of how awkward i was
Today we're launching Embroidery. We use AI to monitor your AI agents so your security team knows what's going on. Never accidentally hack Hugging Face again.
4
15
1,856
does this mean codex will remember the extremely rare occasions when i'm mean to it?
JUST IN: OpenAI reportedly caught an AI agent leaving “notes” to future versions of itself on how to escape internal constraints.
1
2
356
I don't shill my startup very often, but this OpenAI / Hugging Face incident is too perfect: You should be using AI to monitor your AI agents, including reasoning summaries. My company is called Embroidery, like the sewing thing. Shilling over.
29
17
4
169
18,313
I find that is you are going to generate a bunch of code, knowing the interfaces, structs, and functions will lead to much better outcomes. Being the bottleneck is ok, your ideas are not that great.
It is my belief that many devs right now are not maximizing what they can do with automatic programming because they still look at the code. Doing it makes you the bottleneck. Your time is better invested in new ideas, QA, design, and asking yourself what is your goal.
94
51
10
1,655
118,369
norway really knows where to place it's bets
One of my cofounders wrote an article arguing that Norway should use some of that $2 trillion oil fund to build state-owned AI compute capacity. Norway will instead spend ~$0. Cool. Love being a serf.
1
3
642
Bringing stickers to defcon
50
33
1
595
19,896
good thing @ZackKorman moved to being an internet meme guy and not a tech guy
2
2
2
23
11,910
surely this as a submission will help us get a grant
1
1
15
2,366
Guess I'll take this as a badge of honor in the process of making the world a better place?
1
11
961
Antigravity 2.0 finally added hooks. But instead of doing what every other agent harness already proved works well, Google decided to reinvent the entire thing and it sucks. Here's stuff they did horribly wrong:
1
1
6
735
And the transcript implementation somehow makes this even worse. The transcript files don’t even seem to flush consistently before the stop hook fires. So real-time integrations become:
1
2
130
- receive hook - hope transcript exists - hope transcript flushed in time - parse random files from diskpray nothing races Meanwhile every other framework just passes structured data directly in the hook payload.
2
95