I will have to print & frame this tweet :-))

French teachers are certainly much worse paid than German teachers.

whenever I see - "claude/chatgpt has started debugging your browser" - have to give an agent full disk and/or accessibility access I'm convinced we need fundamentally better security primitives for agents to do work as us - and we are living in an awkward intermediate era.

Who said that coding is a solved problem (by AI) they never work on good code base. Change my mind. Micromanage is still necessary or we will end up filled of AI slop till our projects will fail because none, including AI could understand what it does.This after 3 month of Fable.

Craziest thing I saw yet: GLM-5.3 *flash* can exceed Mythos Preview's scores on ExploitBench. Not exactly surprising though. I've been saying for a while that ultimately, TTC scaling destroys categorical capability tiers. Models are now AGI enough to Just keep making progress.

water is transparent only within a very narrow band of the electromagnetic spectrum, so living organisms evolved sensitivity to that band, and that's what we now call "visible light".

I am very uncomfortable about people trying to ascribe religious force or a surrender of human judgment to AI models, and think it is a real safety issue.

I have made a macro to add "Human" and "AI" to papers. Looks good. \usepackage{tcolorbox} \tcbuselibrary{skins,breakable} % Paragraph authorship bars: green Human and blue AI labels in the left margin. % Usage: \begin{humantext} ... \end{humantext} % \begin{aitext} ... \end{aitext} % The body may contain multiple paragraphs, displayed mathematics, or lists. % Blocks break across pages, repeating the vertical label on each segment. % Text keeps its normal width; reserve space in the left margin for the label. % Optional arguments accept tcolorbox keys; redefine HumanBarColor/AIBarColor % after this file is loaded to customize the colors. \ifdefined\gappyParagraphBarsLoaded \else \newcommand{\gappyParagraphBarsLoaded}{} \tcbuselibrary{skins,breakable} \definecolor{HumanBarColor}{HTML}{79CD76} \definecolor{AIBarColor}{HTML}{70B2FF} \tcbset{authorship paragraph/.style 2 args={ enhanced,blanker,breakable,parbox=false, left=8pt,grow to left by=8pt, before skip=\medskipamount,after skip=\medskipamount, borderline west={1.2pt}{0pt}{#1}, overlay={\node[rotate=90,anchor=south east,inner sep=0pt, font=\sffamily\scriptsize\bfseries,text=#1] at ([xshift=-3pt]frame.north west) {#2};} }} \newtcolorbox{humantext}[1][]{authorship paragraph={HumanBarColor}{Human},#1} \newtcolorbox{aitext}[1][]{authorship paragraph={AIBarColor}{AI},#1} \fi

Do journalists do background checks anymore? Constant dedicated profiles to EAs claiming to be "security/safety" experts who are actually 20-something yr olds w/ degrees in policy or economics. Never see dedicated profiles for experts with pedigree who challenge these narratives.

I think consciousness is a red herring. We expand empathy to living beings that almost certainly have no consciousness. If we were to expand empathy to computational processes we created, we end up with playing Lemmings as a bad form of cruelty.

I like Meta's formatting decision to unobtrusively highlight "AI" and "Human" contributions to a paper. Will try to adhere to this practice in the future.

I do wonder whether there is something like "uterus envy", a deep-seated inferiority bevause a man cannot create sentient life in the way a woman can, that inclines toward believing that a mechanical contraption is equivalent to life.

The people that want to believe in "AI consciousness" largely want to believe in their own greatness, and given the unscientific and unfalsifiable nature of "consciousness", there's nothing that will convince them; this is about faith and creating idols, not science.

If I end up having to write a blog entry explaining that a computer program is not a person, I'm going to be very upset

21st century history is the progressive development of the notion that people are automata and vice versa

Si les pays européens ne saisissent pas l’opportunité qu’il y a aujourd’hui avec One Europe One Market🇪🇺, le 28regime et la Savings and Investments Union, l’Union de l’Energie et l’Union de la Connectivité, ils se retrouveront à choisir si il est mieux d’être une colonie 🇺🇸 ou 🇨🇳

This is *actually* my worry about AI. Not that it will “end humanity,” because I’m a believer in logic, not faith, and there are a lot of logical leaps that must be made to cross that very, very wide chasm. But this? Almost guaranteed, and very, very bad.

there is not a single sandbox in the World that's safe from those guys @brucedang @XorNinja @calif_io

The @unpromptedconf II agenda: 101 speakers. 81 sessions, workshops, villages, and more. A first look at ten [un]prompted talks: Bruce Schneier (Harvard Kennedy School; University of Toronto): “Integrous AI Systems Design” @sounilyu (Knostic): “Giving AI Religion” Taesoo Kim (Microsoft): “From DARPA AIxCC to Microsoft FORGE Lab: Lessons Building Hyperscale AI Code Scanning Tools” Matt Suiche @msuiche (Tolmo): “Weightless: Abliteration Without the Weights” Daniel Cuthbert @dcuthbert (Santander Group): “Your sandbox degrades gracefully, what's the wurst that could happen?” @chenxiwang (Rain Capital) and Jasmine Jaksic (NVIDIA): “Agentic AI in the Wild: What Real Deployments Teach Us About Governance and Scale” Tal Waitzenberg and 🛡 Erez Harush (Wiz): “Same Threat, Two Verdicts: Shadow Testing Our SOC Agent” Adrian Wood (Dropbox) and Ads Dawson (Dreadnode): “StealthBench: Can Autonomous Security Agents Do OPSEC Like An Experienced Operator?” Markku-Juhani O. Saarinen (Tampere University): “Math ability has consequences -- AI Cryptanalysis” Kat Traxler @NightmareJS (Vectra AI): “Your Harness Needs a Security Constitution” We’re 26 days out from [un]prompted II, the AI Security Practitioner Conference, where the security community convenes to share what they’ve built, tested, deployed, and what failed.  Please share this lineup with your peers! Stay tuned for more details on the agenda for October 27–29, in San Francisco or online. Register: unpromptedcon.org Thank you to our Sponsors: Knostic, TachTech, Censys @censysio, Ent @Ent_Security, WorkOS @WorkOS, Decibel Partners, Whiterabbit, Zenity @zenitysec, RunSybil @runsybil, AISLE™. Thank you to our Community Partners: SANS Institute @SANSInstitute, Cloud Security Alliance @cloudsa, RSAC @OneRSAC, FIRST @FIRSTdotOrg.