Coercion and Deception in AI-to-AI Management
This article is a summary of an original study by Compassion in Machine Learning (CaML) : Brazilek, J., Chaudhary, M., Lu, Z., & Tidmarsh, M. (2026). Coercion and deception in…
Book Review: The Infinity Machine
It looks like Demis Hassabis is stepping away from Google DeepMind. In honor of his rise and presumed fall, I wrote an essay on the powers and perils of seeing 90% of the future.…
Is it ethical to work on general-purpose robots given the risk of totalitarianism?
One potential risk of developing general-purpose robots is that they could greatly reduce the friction required to establish a totalitarian regime. If robots became physically…
How to be an AI safety research engineer
This is the advice I wish I had when I started trying to become an AI safety research engineer. The Landscape Start by working out which issues you care about. If you don't care…
Is Eval Gaming Downstream of Verbalized Eval Awareness? Not when it's reflexive.
Code and data available at github.com/KieronKretschmar/latent-awareness TL;DR We take two eval-gaming model organisms ( Hua et al.'s (2025) organism and RogueQwen ) and apply…
Taiwan 'Throttles' Mobile Internet For First Time During Annual War Games
Taiwan deliberately throttled mobile internet across the Taichung area during its annual Han Kuang war games, "simulating communication disruption in the event of a Chinese attack…
Defense Against the Deceptive Arts
It's come to my attention that people on Lesswrong are really really really bad at defending themselves against even semi competent deceptions. And worse, they massively…
Microsoft Responds to Outcry After Quietly Installing Beta 'Photos' App on Enterprise Machines
Microsoft's cloud storage app OneDrive got a new Photos app in the worst possible way, reports the blog Neowin . "The app is reportedly showing up even on Windows 11 Enterprise…
Paramount Promises 30 Film-A-Year Contracts After WBD Acquisition
Paramount is making commitments to major theater chains to release 30 films annually for three years if its Warner Bros. deal goes through. According to Deadline, the written…
Spotify Will Label 'AI Persona' Profiles, Exclude Their Music From Recommendations
Spotify will begin labeling AI-generated artist identities with "AI Persona" badges and exclude their music from editorial, algorithmic, and personalized recommendations unless…
Training a Conceptual Reasoning Judge
TL;DR: We fine-tune a judge LLM on our conceptual reasoning dataset to output a critique rating in a single forward pass. This method provides significant uplift in performance on…
Three Supermassive Black Holes Discovered In a Single Galaxy For the First Time
Astronomers using JWST have found three actively feeding supermassive black holes in the distant galaxy J0148-4214, seen as it existed about 1.2 billion years after the Big Bang.…
On Dwarkesh Patel’s Podcast With Ryan Greenblatt
Some podcasts are self-recommending enough that I look to break them down if I have the chance. This, as a debate about recursive self-improvement, was one of those. So here we…
Longtermism Seems Like A Religion
This is a cross post from my blog post . Longtermism is typically framed as an academic philosophy or as a pet project of billionaires, but it’s worth pointing out that, in many…
Demon Safety
(by LemmySmackett ) "Hey man, I haven't seen you in a minute. What are you up to these days?" "Been on that grind, bro. I got a new gig." "Really? You found a job in this dog shit…
Seeing things through in the age of AI
AI is fantastic at prototyping. A quick draft of an essay, a mockup of a website, a demo of a video game, concept art or trailer for a movie, or the core argument of a proof -…