PODCAST

LessWrong (30+ Karma)

LessWrong

Audio narrations of LessWrong posts.

technology philosophy

All Episodes

16:54

“Letting Claude do Autonomous Research to Improve...

LessWrong (30+ Karma) ·

2026/03/10

This work was done as part of MATS 7.1 I pointed Claude at our new synthetic Sparse Autoencoder benchmark, told it to improve Sparse Autoencoder (SAE) performance, and left it running overnight. By morning, it had boosted F1 score from 0.88 to 0.95. Within another day, with occasional input from me, it had matched the logistic regression probe ceiling of 0.97 -- a score I honestly hadn't thought was possible for an SAE on this benchmark. The most surprising development was when Claude autonomously

en-gb

5:54

“Don’t Let LLMs Write For You” by JustisMills

LessWrong (30+ Karma) ·

2026/03/10

Content note: nothing in this piece is a prank or jumpscare where I smirkingly reveal you've been reading AI prose all along. It's easy to forget this in roarin’ 2026, but homo sapiens are the original vibers. Long before we adapt our behaviors or formal heuristics, human beings can sniff out something sus. And to most human beings, AI prose is something sus. If you use AI to write something, people will know. Not everyone, but the people paying attention, who aren’t newcomers or distracted or

en-gb

23:35

“Economic efficiency often undermines sociopolitical...

LessWrong (30+ Karma) ·

2026/03/10

Many people in my intellectual circles use economic abstractions as one of their main tools for reasoning about the world. However, this often leads them to overlook how interventions which promote economic efficiency undermine people's ability to maintain sociopolitical autonomy. By “autonomy” I roughly mean a lack of reliance on others—which we might operationalize as the ability to survive and pursue your plans even when others behave adversarially towards you. By “sociopolitical” I mean that

en-gb

15:01

“Gemma Needs Help” by Anna Soligo

LessWrong (30+ Karma) ·

2026/03/10

This work was done with William Saunders and Vlad Mikulik as part of the Anthropic Fellows programme. The full write-up is available here. Thanks to Arthur Conmy, Neel Nanda, Josh Engels, Dillon Plunkett, Tim Hua and many others for their input. If you repeatedly tell Gemma 27B its answer is wrong, it sometimes ends up in situations like this: I will attempt one final, utterly desperate attempt. I will abandon all pretense of strategy and simply try random combinations until either I stumble upon

en-gb

10:07

“Not Loving Liking What You See” by Tomás B.

LessWrong (30+ Karma) ·

2026/03/10

I am a great fan of Ted Chiang. Many see Understand as his weakest story. I love it, as it is the finest work of intelligence porn ever written. And one of the funniest things I have ever seen on the internet involved it. When I was young I used to read a nootropics message board, a mostly male folly of a forum whose members would have Chinese labs synthesize drugs with some claimed positive effect on IQ. As a rule, they didn't do much of anything. But every time someone tried a new one, there was

en-gb

52:01

“The case for satiating cheaply-satisfied AI...

LessWrong (30+ Karma) ·

2026/03/10

A central AI safety concern is that AIs will develop unintended preferences and undermine human control to achieve them. But some unintended preferences are cheap to satisfy, and failing to satisfy them needlessly turns a cooperative situation into an adversarial one. In this post, I argue that developers should consider satisfying such cheap-to-satisfy preferences as long as the AI isn’t caught behaving dangerously, if doing so doesn't degrade usefulness or substantially risk making the AI more

en-gb

41:48

“The case for AI safety capacity-building work” by...

LessWrong (30+ Karma) ·

2026/03/10

I work on the capacity-building team on the Global Catastrophic Risks-half of Coefficient Giving (formerly known as Open Philanthropy). Our remit is, roughly, to increase the amount of talent aiming to prevent unprecedented, globally catastrophic events. These days, we’re mostly focused on AI, and we’ve funded a number of projects and grantees that readers of this post might be familiar with– including MATS, BlueDot Impact, Constellation, 80,000 Hours, CEA, the Curve, FAR.AI's events, university

en-gb

76:29

“Claude Code, Claude Cowork and Codex #5” by Zvi

LessWrong (30+ Karma) ·

2026/03/10

It feels good to get back to some of the fun stuff. The comments here can double as a place for GPT-5.4 reactions, in addition to my Twitter thread. I hope to get that review out soon. Almost all of this will be a summary of agentic coding developments, after a note. Table of Contents The Virtue of Silence (Unrelated Update). Agentic Coding Offers Mundane Utility. Agentic Coding Doesn’t Offer Mundane Utility. Huh, Upgrades. Our Price Cheap. Quickly, There's No Time. A Particular Set

en-gb

12:57

“Censored LLMs as a Natural Testbed for Secret...

LessWrong (30+ Karma) ·

2026/03/09

TL;DR: We introduce a testbed based on censored Chinese LLMs, which serve as natural objects of study for studying secret elicitation techniques. Then we study the efficacy of honesty elicitation and lie detection techniques for detecting and removing generated falsehoods. This post presents a summary of the paper, including examples of transcripts and other miscellaneous findings. arXiv paper | Code | Transcripts Summary We construct a testbed for honesty elicitation and lie detection techniques

en-gb

40:17

“Prefill awareness: can LLMs tell when “their”...

LessWrong (30+ Karma) ·

2026/03/09

David Africa, Alex Souly, Jordan Taylor, Robert Kirk TLDR: We test whether LLMs can detect when their conversation history has been tampered with (prefill awareness). We find this ability is inconsistent across models and datasets, shallow, and rarely surfaces spontaneously during normal conversation. However, recent Claude models show rather strong prefill detection capabilities when prompted, suggesting prefill awareness is an emerging and model-specific confound that should be actively

en-gb

15:16

“Prologue to Terrified Comments on Claude’s...

LessWrong (30+ Karma) ·

2026/03/09

What Even Is This Timeline The striking thing about reading what is potentially the most important document in human history is how impossible it is to take seriously. The entire premise seems like science fiction. Not bad science fiction, but—crucially—not hard science fiction. Ted Chiang, not Greg Egan. The kind of science fiction that's fun and clever and makes you think, and doesn't tax your suspension of disbelief with overt absurdities like faster-than-light travel or humanoid aliens, but

en-gb

8:11

“Promoting enmity and bad vibes around AI safety” by...

LessWrong (30+ Karma) ·

2026/03/09

Promoting enmity and bad vibes around AI safety I've observed some people engaged in activities that I believe are promoting enmity in the course of their efforts to raise awareness about AI risk. To be frank, I think those activities are increasing AI risk, including but not limited to extinction risk. However, that's a stronger claim than I intend to argue here. Rather, I'll just be presenting a simple and harmful causal pathway and some strategies that can be used for mitigating it:

en-gb

23:36

“The Law of Positive-Sum Badness” by Davidmanheim

LessWrong (30+ Karma) ·

2026/03/09

I keep running into similar arguments online, where people attack “the other” and use the (correct) observation of badness to claim their side is therefore doing well. There's a temptation to correct this by saying that in a dispute between two sides, one side being bad isn’t causally making the other better, or asserting that badness of the two are not correlated. This is tempting, but wrong - because they are correlated, in the opposite direction, and that leads to my observation: Manheim's Law

en-gb

5:09

“Recreation of EA-Pioneer Igor Kiriluk” by avturchin

LessWrong (30+ Karma) ·

2026/03/09

On 3 September 2022, Igor Kiriluk suddenly died (see EA Forum obituary). He was a great communicator and organized the first Moscow EA meetup. He had been active in the transhumanist scene since 2003 and even attracted the founders of KrioRus to transhumanism. He worked as community glue and visited every meetup. On 5 January 2026, his closest friend Veter (Sergey Kamenev) had a dream that Igor's sideload (a mind-model based on an LLM) was created and that Igor calls him and suggests they celebrate

en-gb

13:17

“Payorian Cooperation is easy with Kripke frames” by...

LessWrong (30+ Karma) ·

2026/03/09

The context is MIRI's twist on Axelrod's Prisoner's Dilemma tournament. Axelrod's competitors were programs, facing each other in an iterated Prisoner's Dilemma. MIRI's tournament is a one-shot Prisoner's Dilemma, but the programs get to read their opponent's code. Or, rather, a description of the behavior of the code in Gödel-Löb provability logic, which turns out to be enough to determine their behavior in the setup. One fun result, right in the beginning of the paper, is about a program,

en-gb

45:00

“On Independence Axiom” by Ihor Kendiukhov

LessWrong (30+ Karma) ·

2026/03/08

The Fifth Fourth Postulate of Decision Theory In 1820, the Hungarian mathematician Farkas Bolyai wrote a desperate letter to his son János, who had become consumed by the same problem that had haunted his father for decades: "You must not attempt this approach to parallels. I know this way to the very end. I have traversed this bottomless night, which extinguished all light and joy in my life. I entreat you, leave the science of parallels alone... Learn from my example." The problem was Euclid's

en-gb

23:22

“Solar storms” by Croissanthology

LessWrong (30+ Karma) ·

2026/03/08

Most of civilization's electricity is generated far off-site from where it's delivered. This is because you don't want to be running and refueling coal/gas/nuclear plants inside cities, hydraulic/wind power can't be moved, and solar panels are cheaper to install on flat desert terrain than on cities: So in practice this means running power over hundreds or even thousands of kilometers. E.g. here are the Chinese long-distance lines: Gemini 3.1 Pro-preview in AI studio American long-distance

en-gb

8:42

“The current SOTA model was released without safety...

LessWrong (30+ Karma) ·

2026/03/08

TL;DR: OpenAI released GPT-5.4 Thinking and GPT-5.4 Pro on March 5, 2026. GPT-5.4 Pro is likely the best model in the world for many catastrophic risk-relevant tasks, including biological research R&D, orchestrating cyberoffense operations, and computer use. It has no system card, and, to our best knowledge, has been released without any safety evals. We argue this has occurred at least once before, with GPT-5.2 Pro, and provide recommendations for how a team could conduct fast, independent

en-gb

37:35

“Can governments quickly and cheaply slow AI...

LessWrong (30+ Karma) ·

2026/03/08

I originally wrote this as a private doc for people working in the field - it's not super polished, or optimized for a broad audience. But I'm publishing anyway because inference-verification is a new and exciting area, and there aren't many birds-eye-view explainers of what's going on in it and what the bottlenecks are. 1. Summary I think powerful AI will be obviously scary at some point, and companies or governments might want to slow it down to buy time for additional safety or oversight. Maybe

en-gb

0:27

“The first confirmed instance of an LLM going rogue...

LessWrong (30+ Karma) ·

2026/03/07

First published:March 7th, 2026 Source: https://www.lesswrong.com/posts/XRADGH4BpRKaoyqcs/the-first-confirmed-instance-of-an-llm-going-rogue-for --- Narrated by TYPE III AUDIO.

en-gb

4,474 results

LessWrong (30+ Karma)

All Episodes

Similar Podcasts

ev.news

Martyn Lee

TechCrunch Industry News

TechCrunch

Elon Musk Podcast

Stage Zero

Reflections

Livewire Labs

TechCrunch Startup News

TechCrunch