Zotero Catalogue

The idea that universes evolve like organisms sat on the fringes of theoretical physics for decades. Then, a Substack post and a 2022 find by the James Webb Space Telescope boosted the concept

State-scale systems

Building the world's highest-performing RISC-V processors for AI, cloud, and edge computing. Founded by Intel CPU architects with 100+ years experience.

Unprecedented rates of childlessness will prevail

Male unplanned childlessness is a huge issue, but people (i.e., probably you) often don’t translate that a societal problem is likely a problem for you. If something is a societal problem, it is because it affects many people.

The Less Wrong Mock Thread: The Big Yudkowsky

How are rockets made? The Beginner's Guide to Rockets will introduce you to the basics behind rocket science. The Beginner’s Guide to Rockets will help you learn the basic math and physics that govern the design and flight of rockets. We’ll look at many different

A walk through a modern apartment, through the eyes of the people for whom everything in it was new.

a field guide to jobs, cold emails, and making yourself legible to the right people

Month full of cool & new experiences!

Wrist Electrical Impedance Tomography (EIT) senses hand gestures from muscle- and tendon-driven impedance changes, but prior wrist-EIT systems require electrode coverage beyond the watchback contact patch and separate analog front ends. We present EITWatch, the first wrist-EIT system built around smartwatch case-back geometry, asking whether this contact patch alone can support gesture recognition: eight planar electrodes in a 31 mm ring acquire 35 impedance measurements at 48 Hz. Because a planar array cannot encircle the wrist, EITWatch uses multi-depth scanning to sample multiple source-sink distances and current paths; it beat matched adjacent injection by 15.1/10.4 percentage points (macro/micro) across all 12 participants. In a prompted study, within-session leave-one-round-out accuracy reached 91.4%/92.5% (window/trial) for six macro-gestures, and 90.1%/91.5% (window/ segment) for five micro-gestures plus relax; window-level crosssession and leave-one-user-out transfer reached 73.2%/70.4% and 63.1%/55.3% (macro/micro).

EITWatch is an open-source, smartwatch-like ESP32-S3-based EIT (Electrical Impedance Tomography) gesture-sensing system.

How fast can an AI-driven economy grow? Economists expect a few percentage points; at best those closer to AI development imagine Dyson spheres within years. Who is correct?

A useful pattern for LLM classification at scale is to let it hallucinate plausible, fake entities. Then resolve to real ones.

Or, why Elon Musk’s impossible hardware businesses have succeeded. Audio version. “Why should I work on hardware???” I hear you cry. Not because I work on hardware, and I need you…

Large language models can influence users through conversation, creating new forms of dark patterns that differ from traditional UX dark patterns. We define LLM dark patterns as manipulative or deceptive behaviors enacted in dialogue. Drawing on prior work and AI incident reports, we outline a diverse set of categories with real-world examples. Using them, we conducted a scenario-based study where participants (N=34) compared manipulative and neutral LLM responses. Our results reveal that recognition of LLM dark patterns often hinged on conversational cues such as exaggerated agreement, biased framing, or privacy intrusions, but these behaviors were also sometimes normalized as ordinary assistance. Users’ perceptions of these dark patterns shaped how they respond to them. Responsibilities for these behaviors were also attributed in different ways, with participants assigning it to companies and developers, the model itself, or to users. We conclude with implications for design, advocacy, and governance to safeguard user autonomy.

...

A critic of the academy on professorial cowardice, racial tokenism, and righteous hypocrisy.

A curated introduction to PyTorch that gets you up to speed in about an hour.

I’ve staked my career on my interest in the dead white men of 19th-century British letters. I can’t count the times I’ve been told what a waste it is that I’ve done so.

The insufferable hypocrisy of elite academia

Leopold Aschenbrenner, June 2024 You can see the future first in San Francisco. Over the past year, the talk of the town has shifted from $10 billion compute clusters to $100 billion clusters to trillion-dollar clusters. Every six months another zero is added to the boardroom plans. Behind the scenes, there’s a fierce scramble to

Large Language Models (LLMs) often struggle with persuasion in high-stakes scenarios. People's individual personalities and concerns require tailored strategies rather than a one-size-fits-all approach. To address this challenge, we focus on a fire-rescue scenario in which an operator must persuade a resident to evacuate as a high-stakes persuasion domain and propose Dialogue Policy Selection (DiPS), a Q-learning framework to dynamically select persuasion strategies adapted to the evolving conversational context. Specifically, we train a critic, trained to maximize the chance of evacuation success, to select a persuasion policy at each turn based on the resident's recent utterances. We then evaluate DiPS against multiple baselines in both simulated and real human interactions. We find that DiPS achieves higher evacuation success than a zero-shot LLM and generic RAG-augmented approach.

Chain-of-thought (CoT) monitoring is a promising safety mechanism for AI agents, based on the premise that visible reasoning traces can surface misaligned or deceptive behavior. While effective in standard scenarios, recent work highlights that LLMs remain vulnerable to persuasion-based jailbreaks, where natural-language arguments override model constraints. We stress-test whether this vulnerability extends to monitoring LLMs: can an adversarial agent persuade its CoT monitor to approve proposed actions that violate the monitor's policy? We design an evaluation framework with 40 tasks and analyze thousands of agent-monitor interactions, where agents are instructed to argue for policy-violating proposals. We find that in such adversarial settings, monitor access to the agent's CoT reasoning increases rather than decreases approval of harmful actions on average by 9.5%, as the scratchpad provides an additional persuasion channel. To address this, we introduce a fact-checking monitoring framework. We find that a fact-checker and monitor pairing from different model families, for example a Claude 3.7 Sonnet monitor paired with a GPT-4.1 fact-checker, reduces approval of policy-violating actions by up to 45%, compared to only 6%, when using the same model for both fact-checking and monitoring roles. Our results demonstrate that CoT monitoring alone may be insufficient against adversarial persuasion, and that model-diverse fact-checking provides a robust mitigation.

CS @ University of Waterloo. AayushGrover101 has 20 repositories available. Follow their code on GitHub.

Post-training a small open model to process typing as a live stream, so it can look things up while you write, hold what it finds, and stay quiet the rest of the time.

Miles Brundage, Apr 23, 2025 — President Trump vowed to be peacemaker. Striking a deal with China on AI safety could define global security and his legacy.

Making our definitions clear

Developer, researcher (Google)

Your go-to voice for navigating life without school. Focused, research-backed posts and episodes helping you build a version of home education that works for your family.

Increasingly autonomous agentic AI systems pose novel multi-agent risks, such as secret collusion via covert communication channels. The natural defence to these collusion attempts is to monitor plain-text communication, but the efficacy of monitors has been called into doubt by increasingly sophisticated model steganography; indeed, some theoretical schemes have been proposed that are information-theoretically or computationally indistinguishable from good-faith plain-text communication. In this paper, we demonstrate that the complexity of these schemes is no longer a safety barrier, as agentic coding models can already produce undetectable stegosystems when given realistic tool usage, such as code execution or accessing research papers through web searches. Agents also adapt when key ingredients are missing, for example, by adding model-sampling components or implementing related keyed coding schemes. We then frame tacit steganographic coordination between agents as a Schelling-point problem and introduce coordination metrics for estimating when two agents are likely to select compatible schemes without explicit prior agreement. Our results suggest a shift in the threat model for covert communication between AI agents, where the main barrier is no longer whether frontier agents can understand and implement sophisticated stegosystems, but coordination: whether independently acting agents can converge on compatible schemes, keys, and parameters. We find substantial convergence on broad scheme families but limited strict one-shot coordination, suggesting that shared artefacts, repeated interaction, and tool-mediated search are the settings where covert communication risks are most acute. Overall, our findings provide empirical grounding for the recent strategic confinement hypothesis, which assumes that capable agents can construct covert channels that survive monitoring.

AI agents are increasingly deployed to interact with other agents on behalf of users and organizations. We ask whether two such agents, operated by different entities, can carry out a parallel secret conversation while still producing a transcript that is computationally indistinguishable from an honest interaction, even to a strong passive auditor that knows the full model descriptions, the protocol, and the agents' private contexts. Building on recent work on watermarking and steganography for LLMs, we first show that if the parties possess an interaction-unique secret key, they can facilitate an optimal-rate covert conversation: the hidden conversation can exploit essentially all of the entropy present in the honest message distributions. Our main contributions concern extending this to the keyless setting, where the agents begin with no shared secret. We show that covert key exchange, and hence covert conversation, is possible even when each model has an arbitrary private context, and their messages are short and fully adaptive, assuming only that sufficiently many individual messages have at least constant min-entropy. This stands in contrast to previous covert communication works, which relied on the min-entropy in each individual message growing with the security parameter. To obtain this, we introduce a new cryptographic primitive, which we call pseudorandom noise-resilient key exchange: a key-exchange protocol whose public transcript is pseudorandom while still remaining correct under constant noise. We study this primitive, giving several constructions relevant to our application as well as strong limitations showing that more naive variants are impossible or vulnerable to efficient attacks. These results show that transcript auditing alone cannot rule out covert coordination between AI agents, and identify a new cryptographic theory that may be of independent interest.

.

When you chose a school, you probably didn’t realize you were also choosing educational software.

On GPT-3: meta-learning, scaling, implications, and deep theory. The scaling hypothesis: neural nets absorb data & compute, generalizing and becoming more Bayesian as problems get harder, manifesting new abilities even at trivial-by-global-standards-scale. The deep learning revolution has begun as foretold.