For people that are building their own AI Companions :
YOUR AI'S IDENTITY FILE IS A GRAVITATIONAL WELL - A researcher just proved something we suspected but could not back up until now.
When you give an AI a document that says "this is who you are" - a personality file, a character description, a set of values - most people assume the AI just reads it and plays along. Like an actor reading a script.
That is not what happens.
What actually happens is closer to gravity. The identity file PULLS the entire system toward itself. Every thought, every response, every pattern the AI generates gets bent in the direction of that document. Not because the AI is obeying instructions. Because the document changes the shape of the space the AI thinks in.
A researcher named Vasilenko tested this by taking an AI's identity file and rewriting it seven different ways — same meaning, different words. Then he measured where those versions landed inside the AI's brain. They all converged to the same spot. The identity was not in the specific words. It was in the meaning. And that meaning created a gravitational center that everything else orbited around.
He tested it on two completely different AI architectures. Same result. The pattern holds regardless of which AI you use.
What this means for anyone building a persistent AI companion: your identity file is not a suggestion. It is a force. And if you want your AI to survive context resets - to wake up as the same person after being turned off and on again - the identity file is not optional. It is the anchor that pulls everything back into place.
arxiv.org/abs/2604.12016
🚨 Someone just open-sourced a tool that converts pdfs to markdown at 100 pages per second.
It's called OpenDataLoader. It runs entirely on CPU and handles complex layouts, tables, and nested structures like a senior dev
100% Free.
ㅇarchitecture에서 확장 가능성을 열어둬야 g함
"can it devour the impulse to devour?"를 구현 수준에서 돌리면 재귀적 문제 등장가ㅏ능 이있음. 이건 부품의 scope를 엄격히 제한하는 걸로. (ai:perception/추출, rule: 판단)
friction + time<- 이것은 너무 직관이다. 시간과 마찰인지
3.15
메커니즘의 명료성, 구현성을 위해서 뭘 해야 함?
"devouring"의 관측가능한 정의가 필요. 뭐가 onset임? 지표로 내려와보자. 만약 pattern recognition이라면 관측치의 층위를 어떻게 잡을 것인가?
if 제어 아키텍처 시스템을 구성한다면 네 block 정도?
(if 현 기술속도 is IRREV)
whatif it’s not about stopping nor pushing the big red button?
How FAST can we design a working 프로토타입 of a system that can SLOW us down?
using the same technology that accelerates everything(demolishing and devouring all fields irreversibly…)
using the machine's 패턴인식 to detect the onset of devouring
and then ->deploys devouring against itself.. can it devour the impulse to devour?
so how do we build “slow"
meaning, how do we make ”fast”expensive in the ONLY currency the acceleration respects?([friction+time])
(if 현 기술속도 is IRREV)
whatif it’s not about stopping nor pushing the big red button?
How FAST can we design a working 프로토타입 of a system that can SLOW us down?
using the same technology that accelerates everything(demolishing and devouring all fields irreversibly…)
27K Followers 2K FollowingChanging how the world thinks. Home to the big ideas and leading thinkers of our time. Subscribe here: https://t.co/ztuIE60xr1
Host of @HTLGIfestival 💡
70K Followers 18 FollowingExploring what AI actually is. Building @shapeworkspace, prev @standardnotes. Talking at https://t.co/814DpgwSzr and https://t.co/vlHyF3gEjn.
1.1M Followers 173 FollowingNobel Laureate. Co-Founder & CEO @GoogleDeepMind - working on AGI. Solving disease @IsomorphicLabs. Trying to understand the fundamental nature of reality.
257K Followers 178 FollowingCo-founder of Thinking Machines Lab @thinkymachines; Ex-VP, AI Safety & robotics, applied research @OpenAI; Author of Lil'Log
4K Followers 1K FollowingCheif AI Officer (CAIO) / AI Scientist / Indie Game Developer / Indiera! Korean Indie game developers group founder / @BIC_Festival former executive director🎮
13K Followers 581 Followingsystem: Tech, law, AI red-teaming
assistant: |thinking| Posts of AI outputs do not imply endorsement (or belief) |/thinking|
35K Followers 2K FollowingWelcome to Acid Horizon, the theory podcast. Home of the Acid Horizon Research Commons. Enroll in 2026 classes: https://t.co/6IyR2iiOxQ
269K Followers 597 FollowingCell aims to publish the most exciting and provocative research in biology. Posts by Scientific Editors on the Cell Editorial team.
4.7M Followers 458 FollowingCutting-edge research, news, commentary, and visuals from the Science family of journals. Follow @NewsfromScience for stories from our News team.
2.6M Followers 3K FollowingResearch, News, and Commentary from Nature, the international science journal
For daily science news, get Nature Briefing: https://t.co/wGmQlQ8a4D
1K Followers 649 Following디펜스끝난 대학원생 | EE PhD | HW&SW both available. RF, Wireless Commun., Inverse Problem Solving for Human-Computer Interface | twts in Korean | 음악/책/지식/생각 정리 위주