PODCAST

LessWrong (30+ Karma)

LessWrong

Audio narrations of LessWrong posts.

technology philosophy

All Episodes

24:27

“A Dissent on Honesty” by eva_

LessWrong (30+ Karma) ·

2025/04/15

Context Disney's Tangled (2010) is a great movie. Spoilers if you haven't seen it. The heroine, having been kidnapped at birth and raised in a tower, has never stepped foot outside. It follows, naturally, that she does not own a pair of shoes, and she is barefoot for the entire adventure. The movie contains multiple shots that focus at length on her toes. Things like that can have an outsized influence on a young mind, but that's Disney for you. Anyway. The male romantic lead goes by the name of

en-gb

3:39

“Map of AI Safety v2” by Bryce Robertson, Søren...

LessWrong (30+ Karma) ·

2025/04/15

The original Map of AI Existential Safety became a popular reference tool within the community after its launch in 2023. Based on user feedback, we decided that it was both useful enough and had enough room for improvement that it was worth creating a v2 with better organization, usability, and visual design. Today we’re excited to announce that the new map is live at AISafety.com/map. Similar to the original map, it provides a visual overview of the key organizations, programs, and projects in the

en-gb

5:28

“To be legible, evidence of misalignment probably has...

LessWrong (30+ Karma) ·

2025/04/15

One key hope for mitigating risk from misalignment is inspecting the AI's behavior, noticing that it did something egregiously bad, converting this into legible evidence the AI is seriously misaligned, and then this triggering some strong and useful response (like spending relatively more resources on safety or undeploying this misaligned AI). You might hope that (fancy) internals-based techniques (e.g., ELK methods or interpretability) allow us to legibly incriminate misaligned AIs even in cases

en-gb

20:04

“The Bell Curve of Bad Behavior” by Screwtape

LessWrong (30+ Karma) ·

2025/04/15

There's an implicit model I think many people have in their heads of how everyone else behaves. As George Box is often quoted, “all models are wrong, but some are useful.” I’m going to try and make the implicit model explicit, talk a little about what it would predict, and then talk about why this model might be wrong. Here's the basic idea: A person's behavior falls on a bell curve. 1. Adam is a new employee at Ratburgers, a fast food joint crossed with a statistics bootcamp. It's a great

en-gb

4:15

[Linkpost] “The 4-Minute Mile Effect” by Parker Conley

LessWrong (30+ Karma) ·

2025/04/15

This is a link post. Definition In 1954, Roger Bannister ran the first officially sanctioned sub-4-minute mile: a pivotal record in modern middle-distance running. Before Bannister's record, such a time was considered impossible. Soon after Bannister's record, multiple runners also beat the 4-minute mile. This essay outlines the potential psychological effect behind the above phenomenon — the 4-minute mile effect — and outlines implications for its existence. The 4-minute mile effect describes when

en-gb

4:50

[Linkpost] “Sentinel’s Global Risks Weekly Roundup...

LessWrong (30+ Karma) ·

2025/04/15

This is a link post. Executive summary The Trump administration backtracked from his tariff plan, reducing tariffs from most countries to 10%, but imposing tariffs of 145% on China, which answered with 125% tariffs on US imports and export controls on rare earth metals. The US saw protests against Trump and DOGE, and the US Supreme court ruled that the administration must facilitate the return of an immigrant wrongly deported to El Salvador. OpenAI is slashing the amount of safety testing they will

en-gb

22:45

“Try training token-level probes” by StefanHex

LessWrong (30+ Karma) ·

2025/04/14

Epistemic status: These are results of a brief research sprint and I didn't have time to investigate this in more detail. However, the results were sufficiently surprising that I think it's worth sharing. TL,DR: I train a probe to detect falsehoods on a token-level, i.e. to highlight the specific tokens that make a statement false. It worked surprisingly well on my small toy dataset (~100 samples) and Qwen2 0.5B, after just ~1 day of iteration! Colab link here. Context: I want a probe to tell me

en-gb

21:01

“Frontier AI Models Still Fail at Basic Physical...

LessWrong (30+ Karma) ·

2025/04/14

Dario Amodei, CEO of Anthropic, recently worried about a world where only 30% of jobs become automated, leading to class tensions between the automated and non-automated. Instead, he predicts that nearly all jobs will be automated simultaneously, putting everyone "in the same boat." However, based on my experience spanning AI research (including first author papers at COLM / NeurIPS and attending MATS under Neel Nanda), robotics, and hands-on manufacturing (including machining prototype rocket

en-gb

9:21

“Four Types of Disagreement” by silentbob

LessWrong (30+ Karma) ·

2025/04/14

Epistemic status: a model I find helpful to make sense of disagreements and, sometimes, resolve them. I like to categorize disagreements using four buckets:Facts, values, strategy and labels They don’t represent a perfect partitioning of “disagreement space”, meaning there is some overlap between them and they may not capture all possible disagreements, but they tend to get me pretty far in making sense of debates, in particular when and why they fail. In this post I’ll outline these four

en-gb

21:55

“One-shot steering vectors cause emergent...

LessWrong (30+ Karma) ·

2025/04/14

TL;DR: If we optimize a steering vector to induce a language model to output a single piece of harmful code on a single training example, then applying this vector to unrelated open-ended questions increases the probability that the model yields harmful output. Code for reproducing the results in this project can be found at https://github.com/jacobdunefsky/one-shot-steering-misalignment.Intro Somewhat recently, Betley et al. made the surprising finding that after finetuning an instruction-tuned

en-gb

17:10

“Vestigial reasoning in RL” by Caleb Biddulph

LessWrong (30+ Karma) ·

2025/04/14

TL;DR: I claim that many reasoning patterns that appear in chains-of-thought are not actually used by the model to come to its answer, and can be more accurately thought of as historical artifacts of training. This can be true even for CoTs that are apparently "faithful" to the true reasons for the model's answer. Epistemic status: I'm pretty confident that the model described here is more accurate than my previous understanding. However, I wouldn't be very surprised if parts of this post are

en-gb

9:01

“How I switched careers from software engineer to AI...

LessWrong (30+ Karma) ·

2025/04/13

Thanks to Linda Linsefors for encouraging me to write my story. Although it might not generalize to anyone else, I hope it will help some AI safety newcomers get a clearer picture of how to get into the field. In Summer 2023, I thought that all jobs in AI safety were super competitive and needed qualifications I didn’t have. I was expecting to need a few years to build skills and connections before getting any significantly impactful position. Surprisingly to me, it only took me 3 months after

en-gb

32:25

“College Advice For People Like Me” by henryj

LessWrong (30+ Karma) ·

2025/04/13

I'm graduating from UChicago in around 60 days, and I've been thinking about what I've learned these past four years. I figured I'd write it all down while it's still fresh.This isn't universal advice. It's specifically for people like me (or who want to be like me). High-agency, motivated types who hate having free time.[1] People who'd rather risk making mistakes than risk missing out, who want to control more than they initially think they can, and who are willing to go all-in relatively

en-gb

30:47

“Steelmanning heuristic arguments” by Dmitry Vaintrob

LessWrong (30+ Karma) ·

2025/04/13

Introduction This is a nuanced “I was wrong” post. Something I really like about AI safety and EA/rationalist circles is the ease and positivity in people's approach to being criticised.[1] For all the blowups and stories of representative people in the communities not living up to the stated values, my experience so far has been that the desire to be truth-seeking and to stress-test your cherished beliefs is a real, deeply respected and communally cultured value. This in particular explains my

en-gb

4:13

“Why does LW not put much more focus on AI governance...

LessWrong (30+ Karma) ·

2025/04/12

Epistemic status: Noticing confusion There is little discussion happening on LessWrong with regards to AI governance and outreach. Meanwhile, these efforts could buy us time to figure out technical alignment. And even if we figure out technical alignment, we still have to solve crucial governmental challenges so that totalitarian lock-in or gradual disempowerment don't become the default outcome of deploying aligned AGI. Here's three reasons why we think we might want to shift much more resources

en-gb

25:38

“How training-gamers might function (and win)” by...

LessWrong (30+ Karma) ·

2025/04/12

In this post I lay out a concrete vision of how reward-seekers and schemers might function. I describe the relationship between higher level goals, explicit reasoning, and learned heuristics. I explain why I expect reward-seekers and schemers to dominate proxy-aligned models given sufficiently rich training environments (and sufficient reasoning ability). A key point is that explicit reward seekers can still contain large quantities of learned heuristics (context-specific drives). By viewing these

en-gb

8:48

“Youth Lockout” by Xavi CF

LessWrong (30+ Karma) ·

2025/04/12

Cross-posted from Substack. AI job displacement will affect young people first, disrupting the usual succession of power and locking the next generation out of our institutions. I’m coining the term youth lockout to describe this phenomenon. Youth Lockout We are on track to build AI agents that can independently perform valuable intellectual labour. These agents will directly compete with human workers for roles in the labour market, often offering services at lower cost and greater speeds. In

en-gb

6:22

“Paper” by dynomight

LessWrong (30+ Karma) ·

2025/04/12

Paper is good. Somehow, a blank page and a pen makes the universe open up before you. Why paper has this unique power is a mystery to me, but I think we should all stop trying to resist this reality and just accept it. Also, the world needs way more mundane blogging. So let me offer a few observations about paper. These all seem quite obvious. But it took me years to find them, and they’ve led me to a non-traditional lifestyle, paper-wise. Observation 1: The primary value of paper is to facilitate

en-gb

4:23

“OpenAI Responses API changes models’ behavior” by...

LessWrong (30+ Karma) ·

2025/04/11

Summary OpenAI recently released the Responses API. Most models are available through both the new API and the older Chat Completions API. We expected the models to behave the same across both APIs—especially since OpenAI hasn't indicated any incompatibilities—but that's not what we're seeing. In fact, in some cases, the differences are substantial. We suspect this issue is limited to finetuned models, but we haven’t verified that. We hope this post will help other researchers save time and avoid

en-gb

7:03

“Why do misalignment risks increase as AIs get more...

LessWrong (30+ Karma) ·

2025/04/11

It's generally agreed that as AIs get more capable, risks from misalignment increase. But there are a few different mechanisms by which more capable models are riskier, and distinguishing between those mechanisms is important when estimating the misalignment risk posed at a particular level of capabilities or by a particular model.There are broadly 3 reasons why misalignment risks increase with capabilities: As AIs get more capable, the chance of egregious misalignment increases: AIs with more

en-gb

4,489 results

LessWrong (30+ Karma)

All Episodes

Similar Podcasts

Talking Tech w/ Wayne

Wayne Hunter

The Everything Feed - All Packet Pushers Pods

Packet Pushers

The Good Tech Companies

HackerNoon

Engadget News + Next

Engadget

Cybersecurity Tech Brief By HackerNoon

HackerNoon