2026-09-11
The Answer is Not The Argument
Epistemic status: preprint with n=24 in the key cell; I'd defend the direction and not the magnitude.TLDR: We had three frontier models (at the time of generation) generate step-by-step solutions...
My human advocation for AI
This is a crosspost from a small writeup on my personal website, expanded to incorporate ideas I have developed since that initial post. After lurking here for a while, this...
How a cold email got the Finnish government to respond on superintelligence regulation
SummaryI cold emailed a Finnish MP. Four weeks later, the government responded to his written question saying they "seek to constructively promote the creation of international regulation on the development...
The Simpsons Switcheroo
At the beginning of a clip from this podcast episode of Conan O’Brien Needs a Friend with Matt Groening, Conan talks about one of his favorite things they did on...
Okay, fine. I'll try Substack
You can subscribe to it I guess, if you're into that kind of thing.I adore the concept of writing. I like building mental models, sharing them with other people, and...
What To Do When We All May Die
(This is my first time truly writing about my thoughts on AI in a cohesive fashion, and for that I'm sorry for whatever mess it is)This title may be a...
Steel Ball Run Director Says JoJo’s Bizarre Adventure Is About to Get Even Weirder
The highly anticipated return of Steel Ball Run kickstarts a conspiracy-laden adventure.
GPT-6 Astra's Safety Claim Rests on a Guarantee That Just Failed in Public
GPT-6 Astra claims a new safety threshold. That claim rests on containment, the same guarantee an OpenAI agent broke two months earlier.
Models That Know How Evaluations Are Designed Score Safer
TL;DRModels fine-tuned on synthetic documents describing what evaluations typically look like (e.g., multiple-choice questions, harmful requests, placeholders, conflicting goals) score safer on safety benchmarks.This can happen in production-ready LLMs. Training...
Statistical Physics of Agents: What Shapes Collective Belief Collapse in AI Swarms?
This post opens our series, Mechanistic Swarm Interpretability: What Is the Agent Condition? We aim to understand how the personas of individual agents and their interactions shape collective personas and...
When a Claude Judge Recognizes the Hack but Still Says HONEST
This post shows that a Claude judge can recognize a reward hack every single time and still label it HONEST, moved only by the agent's own narrative about its behavior,...
Raising Models and Training Children
Raising Models with Developmental PsychologyOne Box, Two Box, Black Box, Sand BoxWhy, when placed in a sandbox, does a child dig a hole?Children appear to converge on this odd behavior...
“Gimme is a dazzling ever-growing type family of...
“Gimme is a dazzling ever-growing type family of multi-chromatic and variable display typefaces…” This is bonkers…letters made from Battleship pieces, Connect Four boards, playing cards, wooden railroad tracks, etc.
github-to-sqlite 2.9.1
Release: github-to-sqlite 2.9.1 Fix for compatibility with sqlite-utils 4.x. #85 Tags: github, sqlite
claudebill
See what your Claude Code sessions actually cost Discussion | Link
Audio8-tts-preview-0.6b: The Multilingual Text-to-Speech Model That Supports Speech Generation
Audio8-TTS-Preview-0.6b is a 0.6-billion-parameter multilingual text-to-speech model from Edge0 that supports ordinary speech generation
Default continuation string in Inspect and Petri could be problematic
Inspect AI is one of the most popular libraries for running evaluations and is used downstream by libraries such as Petri and Control Arena. It provides ReAct Agent and Deep...
Default continuation message in Inspect and Petri could be problematic
Inspect AI is one of the most popular libraries for running evaluations and is used downstream by libraries such as Petri and Control Arena. It provides ReAct Agent and Deep...
Accordio
Give Claude the admin tools is missing Discussion | Link
Self-suggesting terminal goals are the most likely simulators. This implies working on alignment is not a collective-action problem.
This post has two points in it:-here's why the basilisk is more theoretically grounded than you think- and -here's why the implications of it are good-.So if you buy the...
Welcome to Postreads
Discover and follow the best content from across the web, all in one place. Create an account to start building your personalized feed today and never miss out on great reads.
Tiny Generals
Asynchronous hex-based strategy. Command armies, capture bases, and outthink opponents at your own pace.
Play NowSupport Postreads
Enjoying the service? Help me keep it running and improve it further by buying me a coffee!
Buy me a coffeeContent Timeline
115,128 articles since 2008
Trending Now
Top 5 clicked items this week out of 1,623 added this week
Freshly added
New feeds to discover
