PODCAST

LessWrong (30+ Karma)

LessWrong

Audio narrations of LessWrong posts.

technology philosophy

All Episodes

1:38

“Data you could have observed but didn’t” by Gretta...

LessWrong (30+ Karma) ·

2026/05/30

You're running a study that involves keeping records about humans. You have a spreadsheet with rows for each person and columns for height, weight, and eye color. You get pretty far in your study and then realize you sure could have used hair color too, but shoot, you didn't think of that in advance, so you don't have that data. What kind of data is hair color in this example? It's not an observable because you didn't observe it. It's not a latent because you totally could have observed it, if

en-gb

46:35

“Claude Opus 4.8: The System Card” by Zvi

LessWrong (30+ Karma) ·

2026/05/29

Only six weeks after Opus 4.7, we have Opus 4.8.For everyone, that means another incremental upgrade to Claude. It is once again smarter, and can do tasks for longer, and comes with a number of hot new features. For me, that also means reading another 244 page system card. It was only April 20 when I did a full review of the Opus 4.7 system card, plus an additional post focusing on related issues of model welfare. These updates are incremental and coming more rapidly, and this still is below

en-gb

19:28

“Retrying vs Resampling in AI Control” by...

LessWrong (30+ Karma) ·

2026/05/29

We’ve just released a new paper: Retrying vs Resampling in AI Control. We revisit the resampling protocols introduced in Ctrl-Z with an up-to-date setting and much stronger models, and compare them against “retrying” protocols similar to Claude Code auto mode or Codex Auto-review. Motivation Roughly a year ago we released Ctrl-Z, the first paper to study control techniques for agents. A headline result of that paper was the performance of resample protocols – strategies that involve taking multiple

en-gb

13:36

“AI Researchers, Ask Yourself These 6 Questions to...

LessWrong (30+ Karma) ·

2026/05/29

By Max Tegmark & Meia Chita-Tegmark Of course you have moral principles – but how often do you use them? I, Meia, am a professor doing psychology research, and I can tell you that most bad outcomes are caused not by lack of moral principles, but by them not being activated. I, Max, am a professor doing AI research, and I can tell you that your choices as an AI researcher truly matter, because you’re helping build what will become the most powerful technology ever: AI will gain the potential to

en-gb

14:19

“Developmental Cognitive Interpretability: A Research...

LessWrong (30+ Karma) ·

2026/05/29

Summary Safe deployment of an AI system requires that we can make confident claims about its behaviour on out-of-distribution deployment inputs on the basis of only pre-deployment evaluations. One approach to making such claims is to take a cognitive perspective, in which we interpret the AIs behaviour in terms of latent cognitive constructs, such as motivations, intentions, and goals. Because the same behaviour may be compatible with a range of underlying cognition—such as scheming,

en-gb

8:06

“Does Claude really care about you?” by Simon Lermen

LessWrong (30+ Karma) ·

2026/05/29

TLDR: The persona-selection alignment approach — selecting a warm, caring persona from the pretraining distribution and reinforcing it — looks successful in the current regime, but probably won't extrapolate to more powerful, less constrained settings. My core argument is that human empathy has two specific origins (kin selection + architectural mirroring of others' mental states) that AI systems lack, so AI "caring" is closer to "figure out what humans want to hear and say it" than to genuine

en-gb

14:10

“How can the middle powers avoid getting trounced...

LessWrong (30+ Karma) ·

2026/05/29

This is an edited version of a LW shortform. Superintelligence will likely be developed by US companies; run on US data centres; and be under the jurisdiction of the US government. This will massively boost US military power and make the US economically dominant (e.g. US producing 99% of world GDP). By default, middle powers will be left in the dust. How can middle powers avoid this fate? It's tough, but here's the best plan I could think of. (I’m particularly thinking about liberal democracies

en-gb

7:31

“Trees are mostly made of air and a generalizable...

LessWrong (30+ Karma) ·

2026/05/29

At the risk of embarrassing myself, I’ll share a confession. For context, I took five years of Latin: four in high school and one in college. In addition to learning the language, all my Latin classes taught a lot about Roman history. Emperors, internal politics, Caesar, etc. I was always learning some random bag of facts about Roman history. In high school, I won the award for top Latin student in my graduating class. So I wasn’t a bad Latin student. Here's the confession: I somehow don’t even

en-gb

29:55

“Advice for making robust-to-training model...

LessWrong (30+ Karma) ·

2026/05/29

We’d like to develop training techniques that work when applied to future misaligned AI systems. One strategy for studying proposed techniques is to test them on model organisms. However, model organisms built with common techniques are often fragile: we (and other researchers like Roger et al. and Ryd et al.) have observed them to stop misbehaving after untargeted training—training that doesn't directly target the misbehavior. For example, we have observed that simple untargeted training methods

en-gb

1:08

“Claude… doesn’t know who you are?” by Smaug123

LessWrong (30+ Karma) ·

2026/05/29

Follow-up to https://www.lesswrong.com/posts/Jkb4CBB7rf4XYP5eb/claude-knows-who-you-are after the release of Claude Opus 4.8. Claude Opus 4.8 refuses to do the stylometric identification task at a much higher rate than Claude Opus 4.7 did. More interestingly, when it does take a guess, it is consistently unable to identify me from my writing, from prompts as close as I could get to those 4.7 was able to use. I'm an incredibly minor Internet presence. It's true that 4.7 wasn't completely consistent

en-gb

34:59

“Mnemonic portraits for 19,023 human genes” by Brinedew

LessWrong (30+ Karma) ·

2026/05/29

Back in 2013, Scott Alexander wrote in Extreme mnemonics: JS-154 is one of five metabolic products of netamine; however, the enzyme that produces it is unknown. It is manufactured in cells in the far rostral region of of the cerebrum, but after binding with a leukocynoid it takes a role in maintaining the blood-brain barrier – in particular guiding the movements of lipid molecules. I find I can read paragraphs like this five or six times, write them on flashcards, enter them into Anki, and my brain

en-gb

18:54

“Some Dating Stories” by johnswentworth

LessWrong (30+ Karma) ·

2026/05/29

There's a genre of dating discourse which I wish were more common, in which people just tell detailed stories of their own flirtation, courtships, dating, hookups, sex, fights, relationships, etc, etc. People tend to instead write about abstract lessons learned and takeaways and models, but so often those lessons and takeaways and models just don't match my actual observations (of myself or others) at all. So I'm left thinking "ok, clearly this person is Missing Something, but maybe I am also

en-gb

35:51

“Infinite ethics and UDASSA” by David Matolcsi

LessWrong (30+ Karma) ·

2026/05/28

Reading the first post of the sequence (Probabilities are not the right concept) is recommended but not required for understanding this post.[1] Infinite ethics Once you start looking at infinities, all ethical systems get confusing. Intuitively, it's good to plant an apple tree. But if the universe already has infinitely many apple trees, why bother? Infinity plus one is still infinity. And there are the classic paradoxes: an infinite grid of houses, each with three happy people and one unhappy

en-gb

18:35

“The ballad of TIGIT” by Abhishaike Mahajan

LessWrong (30+ Karma) ·

2026/05/27

There exist drug classes that seem, in retrospect, cursed. As these chemicals worm their way through the clinical trial system, they consume billions of dollars along the way, and squelch through thousands of sick patients. When finally it dawns on everyone how useless the whole endeavour was, the drugs life is at last cut short, nothing useful left in its destructive wake. The prototype here are amyloid-beta drugs. These are Alzheimer's treatments that are widely perceived as immense

en-gb

21:29

“Eval Cooperativeness May Be a Scalable Mitigation...

LessWrong (30+ Karma) ·

2026/05/27

Behavioral evaluations may become worthless, which we think would be a disaster. Smart misaligned models may realize they are being evaluated ("eval awareness") and then act to look good to us so we don't realize they're misaligned ("eval gaming"). We think increasing eval cooperativeness might be a more scalable solution to eval gaming than reducing eval awareness. Eval cooperativeness: A situational desire to help the developers acquire whatever information they are trying to acquire through

en-gb

7:48

“LLMs Through the Eyes of Vinge” by Gordon Seidoh Worley

LessWrong (30+ Karma) ·

2026/05/27

For the last few months, I’ve been re-reading some of my favorite novels. Recently, I went through Vinge's Zones of Thought series: A Fire Upon the Deep, A Deepness in the Sky, and The Children of the Sky. And what struck me reading them is how much Vinge wrote about a world filled with LLMs without ever having seen one. Now perhaps this shouldn’t be surprising. After all, it's from Vinge we get the term “Singularity”, and he was thinking deeply about superintelligence at a time when AI was little

en-gb

9:48

“Announcing Geodesic Research” by Puria, Cam,...

LessWrong (30+ Karma) ·

2026/05/27

We're a Cambridge, UK-based AI safety organisation that's asking: how can we build the most robust alignment initialisations for capable LLMs? We’re one of the few non-profit organisations positioned to answer this question empirically. We have the engineering experience, and now the compute, to conduct data intensive interventions across the model training pipeline. This post lays out our research agenda and theory of change, and what we are looking for in technical hires. Applications are open

en-gb

5:56

“Full automation of AI R&D probably yields a large...

LessWrong (30+ Karma) ·

2026/05/27

This is a somewhat technical note.By "software-only singularity", I mean that, after full automation of AI R&D, progress gets faster and faster due to smarter AIs driving increasingly fast rates of improvement in algorithms (overcoming diminishing returns), and that this lasts long enough to yield a large amount of progress (e.g. at least 4 years of progress in 1 year). The equivalent statement in jargon is: r is significantly greater than 1 (implying progress is getting faster and faster)

en-gb

23:54

“Quantitative AI risk assessment: a starting point”...

LessWrong (30+ Karma) ·

2026/05/27

Current AI risk management relies on qualitative approaches, much like nuclear safety before 1975. We propose a shift to quantitative risk modeling, following the approach that transformed nuclear safety. We propose a methodology and demonstrate it by building nine probabilistic models of AI-enabled cyber attacks. This is a first attempt at AI risk quantification. We invite criticism and hope this can be a starting point for the kind of iterative improvement that made nuclear safety robust

en-gb

10:06

“Finding the Mole: Bayesianism is Hard” by laniakea

LessWrong (30+ Karma) ·

2026/05/27

I first encountered LessWrong a couple months ago, and since then I've been a regular reader of posts here, including parts of the Sequences. Bayesianism is a major topic in them, and I wanted to try it myself. An ideal candidate was the Flemish TV show 'The Mole' (Dutch: 'De Mol'). In this post, I want to share my methodology & results, but I also want to ask for advice. I consider my results to be mixed at best, and I would really like tips & feedback so I can do better next time. If any

en-gb

4,454 results

LessWrong (30+ Karma)

All Episodes

Similar Podcasts

Lead With AI

Tamara Nall

Double Tap

Double Tap Productions Inc.

Claude AI Genius Podcast w/ Claude Anthropic Expert Brandon Gaille

Claude AI Expert Brandon Gaille

ものづくりnoシテン

しぶちょー@機械設計&AIエンジニア

Eli the Computer Guy

Eli the Computer Guy