Denný feed LIVE
Ranný briefing. Živé aktualizácie. Trhové signály. Všetko na jednom mieste.
Aktualizované 9. marca 2026 · 23:28
TL;DR: We think allowing frontier AI models to be used for mass domestic surveillance and to operate as fully autonomous weapons creates significant risks of emergent misalignment. For those somehow unaware, the Department of War and Anthropic have had a recent dispute over the u
LessWrong →Cortical Labs grew 200,000 human neurons in a petri dish, connected them to silicon, and taught them to play DOOM. The CL1 is a commercially available biological computing unit — neurons as hardware. The cells learn via feedback loops, not code.
YouTube / Cortical Labs →A new macOS-native tool for isolating and sandboxing local AI agents. 642 HN points in 16 hours — that's a signal. The community is building containment infrastructure for agents running on personal machines. The very existence of this tool confirms that autonomous agents are now sufficiently common on laptops to warrant their own security tooling category.
Hacker News →The argument: Knuth's literate programming (code written for humans to read, machines secondary) becomes newly relevant when AI agents write the code. If agents generate the implementation, the human's job shifts entirely to intent documentation — which is exactly what literate programming was designed for.
Hacker News →Zack M. Davis writing a "prologue to terrified comments" on Anthropic's updated model spec/constitution. Not viral, but the title is a signal — and it tracks directly against Andrew Critch's same-day post on "promoting enmity around AI safety" (32✦). Two senior voices from the alignment community publicly anxious about governance direction on the same morning.
LessWrong →The 9th Circuit ruled that Terms of Service can be modified by email notification, and continued use of a product implies consent to the new terms. This has implications for every AI product, every educational platform, every data agreement students and teachers click through.
US 9th Circuit Court of Appeals →Sticky inflation -> higher (real) rates needed -> pain for hyper-financialized economies.
A lot of venture funded CEOs are sending emails like this; inspired by Elon and taking drastic action.
Last week, the Defense Department failed its fifth audit, unable to account for more than 61% its assets, worth about $1.8 trillion.
So for $20/month, you get 2 TB storage, access to all Gemini models (including Deep Research with likely higher limits than OpenAI), and integration with your Gmail and
I like to train large deep neural nets. Previously Director of AI @ Tesla, founding team @ OpenAI, PhD @ Stanford.