PODCAST

LessWrong (30+ Karma)

LessWrong

Audio narrations of LessWrong posts.

technology philosophy

All Episodes

17:11

“Latent Introspection (and other open-source...

LessWrong (30+ Karma) ·

2026/03/24

@vgel, Martin Vanek, @Raymond Douglas, @Jan_Kulveit — ACS Research, CTS, Charles University --- Paper | Code | Earlier post | Twitter thread | Bluesky thread --- Last year, Lindsey demonstrated that Claude models can detect when concepts have been injected into their activations using steering vectors, which Lindsey uses as a proxy test for introspection. If models can detect when concepts have been injected into their activations, it stands to reason they can access their own, naturally-occurring

en-gb

13:23

“The Fourth World” by Linch

LessWrong (30+ Karma) ·

2026/03/24

Is consciousness the last moral world? Imagine trying to explain to a virus why suffering matters. A virus is a simple self-replicating molecule: unsophisticated and arguably not even alive. It has no experience. It just copies itself according to chemical laws. From its “perspective” (it doesn’t have one), the universe is just physics: particles following rules. If you could somehow tell it that certain arrangements of matter are good and others are bad, it wouldn’t disagree with you. It does not

en-gb

10:38

“My cost-effectiveness unit” by Zach Stein-Perlman

LessWrong (30+ Karma) ·

2026/03/24

It feels like the grantmaking around me is only partially moneyball-pilled, or it's only somewhat competent at moneyball. There's alpha in putting numbers on stuff, if you can do it right. Five months ago I wanted to compare a bunch of different kinds of donation opportunities. I needed a universal unit of cost-effectiveness, and for that I needed a unit of goodness. Consider a value scale where "EV of the multiverse" is 100 and "EV of the multiverse, in the counterfactual where the Sun goes

en-gb

12:33

“The AIXI perspective on AI Safety” by Cole Wyeth

LessWrong (30+ Karma) ·

2026/03/24

Epistemic status: While I am specialized in this topic, my career incentivizes may bias me towards a positive assessment of AIXI theory. I am also discussing something that is still a bit speculative, since we do not yet have ASI. While basic knowledge of AIXI is the only strict prerequisite, I suggest reading cognitive tech from AIT before this post for context. AIXI is often used as a negative example by agent foundations (AF) researchers who disagree with its conceptual framework, which many

en-gb

39:16

“Measuring and improving coding audit realism with...

LessWrong (30+ Karma) ·

2026/03/24

TL;DR We study realism win rate, a metric for measuring how distinguishable Petri audit transcripts are from real deployment interactions. We use it to evaluate the effect of giving the auditor real deployment resources (system prompts, tool definitions, and codebases). Providing these resources to the auditor increases the average realism win rate from 4.6% to 32.8% for reward hacking audits, and doesn’t significantly change hack rate. 💻Code Research done as part of the Anthropic Fellows Program

en-gb

12:38

“Ablating Split Personality Training” by OscarGilg

LessWrong (30+ Karma) ·

2026/03/24

I was part of the SPAR team that worked on Split Personality Training: Revealing Latent Knowledge Through Alternate Personalities. I ran some follow-ups. The views and mistakes here are my own. Thanks to William Wale and Florian Dietz for useful comments on the draft TLDR: I ablated the key components of Split-Personality training to understand which parts are load-bearing. (1) Simple user follow-up prompts work just as well as the split-personality framing, and training with this format converges

en-gb

32:19

“AI character is a big deal” by wdmacaskill, Tom...

LessWrong (30+ Karma) ·

2026/03/24

0. Intro Due to Claude's Constitution and OpenAI's model spec, the issue of AI character has started getting more attention, particularly concerning whether we want AI systems to be “obedient” or “ethical”.[1] But we think it's still not nearly enough. AI character (e.g. how obedient, honest, cooperative, or altruistic AIs are, and in what circumstances) will have a big effect on society, and on how well the future goes. We think that figuring out what characters AI systems should have, and getting

en-gb

17:27

“The Federal AI Policy Framework: An Improvement, But...

LessWrong (30+ Karma) ·

2026/03/24

The Federal AI Policy Framework has been released. Well, it is a four page outline. Mostly it just reiterates existing such outlines. But that is four more pages than we had previously. It includes the beginnings of actual policy proposals, some of which are highly welcome and actively good.Perhaps most importantly, it affirms that we are a Republic in which the way we Do Policy is we pass a law through Congress specifying what we do, and that we need to actually Do Policy alongside trying to ban

en-gb

11:54

“Some things I noticed while LARPing as a grantmaker”...

LessWrong (30+ Karma) ·

2026/03/23

Written to a new grantmaker. Mostvalue comes from finding/creating projects many times your bar, rather than discriminating between opportunities around your bar. If you find/create a new opportunity to donate $1M at 10x your bar (and cause it to get $1M, which would otherwise be donated to a 1x thing), you generate $9M of value (at your bar).[1]If you cause a $1M at 1.5xopportunity to get funded or a $1M at 0.5xopportunity to not get funded, you generate $500K of value. The former is 18 times as

en-gb

9:24

“Understanding when and why agents scheme” by Mia...

LessWrong (30+ Karma) ·

2026/03/23

TL;DR To understanding the conditions under which LLM agents engage in scheming behavior, we develop a framework that decomposes the decision to scheme into agent factors (model, system prompt, tool access) and environmental factors (stakes, oversight, outcome influence) We systematically vary these factors in four realistic settings, each with scheming opportunities for agents that pursue instrumentally convergent goals such as self-preservation, resource acquisition, and goal-guarding We find

en-gb

10:18

“Attend the 2026 Reproductive Frontiers Summit, June...

LessWrong (30+ Karma) ·

2026/03/23

We’ll be hosting the 2026 Reproductive Frontiers Summit at Lighthaven in Berkeley, CA, on June 16—18. Come join us if you want to learn, connect, think, and coordinate about the future of germline engineering technology. Very early bird tickets are available now until the end of March. Who will be there? Our lineup of speakers includes experts in the areas of polygenic prediction, embryo gene editing, in vitro gametogenesis, artificial wombs, ethics and regulation for advanced reproductive

en-gb

13:38

“Is fever a symptom of glycine deficiency?” by Benquo

LessWrong (30+ Karma) ·

2026/03/22

A 2022 LessWrong post on orexin and the quest for more waking hours argues that orexin agonists could safely reduce human sleep needs, pointing to short-sleeper gene mutations that increase orexin production and to cavefish that evolved heightened orexin sensitivity alongside an 80% reduction in sleep. Several commenters discussed clinical trials, embryo selection, and the evolutionary puzzle of why short-sleeper genes haven't spread. I thought the whole approach was backwards, and left a

en-gb

5:48

“My Most Costly Delusion” by Ihor Kendiukhov

LessWrong (30+ Karma) ·

2026/03/22

Suppose there is a fire in a nearby house. Suppose there are competent firefighters in your town: fast, professional, well-equipped. They are expected to arrive in 2–3 minutes. In that situation, unless something very extraordinary happens, it would indeed be an act of great arrogance and even utter insanity to go into the fire yourself in the hope of "rescuing" someone or something. The most likely outcome would be that you would find yourself among those who need to be rescued. But the calculus

en-gb

1:06

“China declares AGI development to be a part of...

LessWrong (30+ Karma) ·

2026/03/22

The CCP writes in its 15th 5-year plan that it will. Encourage innovation in multimodal, agentic, embodied, and swarm intelligence technologies, and explore development paths for general artificial intelligence. This is translated from the original: 鼓励多模态、智能体、具身智能、群体智能等技术创新，探索通用人工智能发展路径。 Source: https://www.spp.gov.cn/spp/tt/202603/t20260313_723954.shtml The English-language commentary I found does not have much more to say about this, e.g.:

en-gb

13:41

“Pre-Review of Toy Story 5” by Raemon

LessWrong (30+ Karma) ·

2026/03/22

I am the second most spoiler-averse person I know. (Maybe tied for 2nd with a couple other people?). I once was considering going to an immersive experience, and someone told me the company that ran the experience, and this was enough for me to derive an important twist that'd happen to me in the first few minutes, and I was like "augh that was a spoiler!!!" and they were like "!??". I then went to the experience, and indeed, it was a lot worse than it would have been if I had gotten to be

en-gb

9:34

“Does Hebrew Have Verbs?” by Benquo

LessWrong (30+ Karma) ·

2026/03/22

Spinoza's Compendium of Hebrew Grammar (1677, posthumous, unfinished) contains a claim that scholars have been misreading for centuries. He says that all Hebrew words, except a few particles, are nouns. The standard scholarly reaction is that this is either a metaphysical imposition (projecting his monistic ontology onto grammar) or a terminological trick (defining "noun" so broadly it's vacuous). Both reactions are wrong, and they're wrong for the same reason: they import Greek and Latin

en-gb

7:47

“Untrusted Monitoring is Default; Trusted Monitoring...

LessWrong (30+ Karma) ·

2026/03/21

These views are my own and not necessarily representative of those of any colleagues with whom I have worked on AI control. TL;DR: It's much cheaper and quicker to just throw some honeypots at your monitor models than to robustly prove trustedness for every model you want to use. Therefore I think the most likely future involves untrusted monitoring with some monitor validation as a default path. This post talks about two different ways of monitoring AI systems, trusted monitoring, and untrusted

en-gb

6:36

“China Derangement Syndrome” by Arjun Panickssery

LessWrong (30+ Karma) ·

2026/03/21

Often I see people claim it's essential for America to win the AI race against China (in whatever sense) for reasons like these: “What is the reason we want America to win the AI race? It'sbecause we want to make sure free open societies can defend themselves”(Alec Stapp) “We should seek to win the race to global AI technological superiority and ensure that China does not…to ensure that our way of life is not displaced by the much darker Chinese vision“ (Marc Andreessen) “Will it be one in which

en-gb

14:53

“Contrastive features elicit different perturbation...

LessWrong (30+ Karma) ·

2026/03/21

Note: This is a research update sharing preliminary results as part of ongoing work. Figure 1: Contrastive (difference-of-means, English→Mandarin) feature directions elicit a downstream response at much smaller perturbation magnitudes than SAE directions, which behave similarly to random directions. This holds across multiple models and experimental setups. Summary & Main Results Understanding how concepts are represented in LLM internals would be extremely useful for AI safety (generally

en-gb

11:07

“Confusion around the term reward hacking” by...

LessWrong (30+ Karma) ·

2026/03/21

Summary: "Reward hacking" commonly refers to two different phenomena: misspecified-reward exploitation, where RL reinforces undesired behaviors that score highly under the reward function, and task gaming, where models cheat on tasks specified to them in-context. While these often coincide, they can come apart, require distinct interventions, and lead to distinct threat models. Using a blanket term for both can obscure this. Distinct phenomena qualify as reward hacking The term[1] commonly points

en-gb

3,823 results

LessWrong (30+ Karma)

All Episodes

Similar Podcasts

Lead With AI

Tamara Nall

Linux in the Ham Shack

Black Sparrow Media

Data Engineering Podcast

Tobias Macey

Eye On A.I.

Craig S. Smith

Reflections

Livewire Labs