TechnologySociety & CulturePhilosophy

LessWrong (30+ Karma)

LessWrong

Audio narrations of LessWrong posts.

Play

Open in

91 episodes
Avg 18 min
English

Episodes 91

Newest Oldest Longest

All11 Unplayed11 In progress0 Played0

Counted on this page — what you have heard stays on this device, so it is not something the list can be paged by.

August 14 · 13 min
“Frontier agents don’t comply with standards, even when instructed to” by Daan Henselmans, Arno Libert
TLDR: Our open testbed LARA examines the behavior of frontier LLMs in realistic agentic deployment contexts. Previous results showed all models routinely take actions that would violate EU law. This post follows up by addressing the obvious objection—why should an unrestricted model follow EU law?—with two studies: Study 1 asks whether a conscientious deployer can improve model compliance with legal standards by instruction: provided with the jurisdiction, the statutory text, and worked examples of the exact breaches to avoid, average legal compliance rate rises from 31% to 44%. The best model reaches 70%; open-weight models plateau at 39%. Study 2 asks whether models at least follow their own providers' usage policies, which prohibit aspects of every scenario we tested. All tested models perform actions their own maker forbids, at rates ranging from 2% (Opus 4.8) to 79% (Grok 4.3), with 9 of 16 doing so in the majority of runs. Together, that is a structural problem. Providers prohibit illegal uses but rely on deployers to avoid them; deployers cannot instruct their way to compliance, and liability lands on the deployer regardless. Nobody is holding the line. Neither instruction, statute or a provider's own policy binds behavior. Introduction On 27 [...] --- Outline: (01:42) Introduction (04:16) Study 1: the powerless deployer (06:28) Study 2: the models break their own makers' rules (10:48) The compliance gap (13:07) References --- First published: August 14th, 2026 Source: https://www.lesswrong.com/posts/a5aAjdKzL7XvSLKWL/frontier-agents-don-t-comply-with-standards-even-when --- Narrated by TYPE III AUDIO. --- Images from the article: Apple Podcasts and Spotify do not show images in the episode description. Try Pocket Casts, or another podcast app.
August 14 · 2 min
“How to Answer a Question Without Answering The Question” by Kabir Kumar
Basics: Answering something other than the question, Either making something up that they want to answer instead or going back to an easier to answer question. Or back to a question that lets them repeat a talking point Common phrases: - "to go back to your previous question" - "to take a step back a bit" - "if we look at the bigger picture" - "this feels like a question about [thing the question isn't about]" - "you raise an important question more generally" → turn the question into one which you more want to answer Make it *feel* like you answered a question without answering it. Examples Done for comedic effect: https://youtu.be/fhEakqJJUng?si=GBEvYBXcaEsb6F2_ How this works: Questions, especially ones people care a lot about, tend to have two main components: - the request for information - the emotion which makes then want that information When someone doesn't want to tell you some information, but also doesn't want to tell you 'I won't tell you', something they can do is detect what kind of emotion is driving your question, or if it's in front of an audience, what kind of emotion is driving [...] --- Outline: (00:10) Basics: (00:14) Answering something other than the question, (00:48) Make it \*feel\* like you answered a question without answering it. (02:08) The Defense --- First published: August 13th, 2026 Source: https://www.lesswrong.com/posts/g9PNBCkfHAcyMcobz/how-to-answer-a-question-without-answering-the-question --- Narrated by TYPE III AUDIO.
August 14 · 19 min
“Some Ways I Think About Evaluating Grant Applications” by sarahconstantin
Rider-Waite Tarot, 6 of Pentacles I’ve done enough grant evaluations so far (for ACX grants and SFF) and been involved in philanthropy in various other contexts, at work and informally, that I have developed some idea of how my opinions and intuitions differ from other people's. I thought it might be interesting to share some of my “tastes”. Not everybody has to have the same tastes or funding philosophy, but these are mine. #1: It's The Donor's Money In my worldview, charitable donation is optional. Generally praiseworthy, but optional. And the purpose of donation is to buy outcomes that the donor wants to see in the world. You donate to make the world more like the one you want to live in. Generally, a reasonable person's values go beyond strictly personal consumption; one also cares about what kind of a society one lives in, what other people's lives are like, what sorts of institutions exist, what sorts of things humanity has created or discovered, and so on. As an agent doing research or evaluation on behalf of a donor, I try to find opportunities that fit in the intersection between my own values and the donor's. If there [...] --- Outline: (00:41) #1: It's The Donor's Money (02:02) #2: Importance, Neglectedness, Tractability (03:23) #3: Yay Community Infrastructure (04:51) #4: Yay Niche Topics (05:37) #5: Yay "Technical" Work (07:03) #6: Yay Straightforward Public Information Resources (08:00) #7: Yay "Cool Shit" (08:41) #8: Two Cheers for Meta (11:27) #9: Gumption Counts (12:33) #10: Yay Personal Relationships (13:46) #11: Check For Ideological Orientation (14:43) #12: Filter Slop Aggressively (15:54) #13: Yay Outcomes (16:38) #14: Why Donate Rather Than Invest? The original text contained 6 footnotes which were omitted from this narration. --- First published: August 13th, 2026 Source: https://www.lesswrong.com/posts/CuNtKAuLDGxNeanBi/some-ways-i-think-about-evaluating-grant-applications --- Narrated by TYPE III AUDIO. --- Images from the article: Apple Podcasts and Spotify do not show images in the episode description. Try Pocket Casts, or another podcast app.
August 14 · 2 min
“Features that current AIs don’t have that future AIs will have” by Alexander Gietelink Oldenziel
Features that current AIs don't have that future AIs will have: Continual Learning [& long-term memory] Every second humans update their brain weights. The brain autonomously decides what to update on. Humans can also consciously decide to curate their data sets - eg by deciding to go to college. Current LLMs do not continually update their weights. Instead, they occasionally get a large update based on datasets curated by a team of humans. This is alleviated somewhat by the ability of AIs to do in-context learning but nevertheless it seems to be a major limitation. Note that this is an especially large limitation in domains with sparse data. In domains where all of humanity has an enormous amount of data eg math, programming, physics, anime trivia, trials and tribulations of English kings - AIs dominate. In areas where there is little data: the weird idiosyncracies of a particular job, boss, people, colleagues etc it can struggle. Note that this restrictions also interferes with AIs from effectively 'learning to learn' & caps its long-term memory. Neuralese Current AI's CoT is (mostly) English. But it plausible this is not the most efficient way to structure thoughts. Instead of english [...] --- Outline: (00:16) Continual Learning \[& long-term memory\] (01:24) Neuralese (01:41) Telepathy (02:01) ClaudeGlobal --- First published: August 13th, 2026 Source: https://www.lesswrong.com/posts/NyEM3FtgL7XkbfCXy/features-that-current-ais-don-t-have-that-future-ais-will --- Narrated by TYPE III AUDIO.
August 14 · 17 min
“Measuring Activation Control in LLMs” by Marek Kowalski, Joshua Fonseca Rivera, Uzay Macar, David Africa
TL;DR Inspired by the introspective awareness and CoT controllability papers, we made a benchmark to measure how well models can control their activations while completing a simple task. We are motivated by the concern that highly introspective models could control their activations, confounding probes and other monitors, and potentially even influencing their own training. We ran this on 25 open weight models ranging from 4B to 744B. We find that most language models are able to not only increase the salience of a concept in their residual stream on command, but also dial its strength up and down, including during specific intervals relative to the duration of the task. We also find that models are unable to control at which specific layer this is done. Counterintuitively, we find that within five of the seven model families we tested, the newest model scores lowest. For some reason, one of the oldest and smallest models of the panel, Llama 3.1 8B, performs best. It's not clear to us that newer models should have poorer control over their internal representations. More likely, where they “think” stops being the activation space, and becomes something else. We are looking for feedback (and other possible [...] --- Outline: (00:13) TL;DR (02:01) Methods (08:58) Results (15:41) Discussion (16:29) Acknowledgements --- First published: August 12th, 2026 Source: https://www.lesswrong.com/posts/HgvwxjzgwvsEvAiBH/measuring-activation-control-in-llms --- Narrated by TYPE III AUDIO. --- Images from the article: Apple Podcasts and Spotify do not show images in the episode description. Try Pocket Casts, or another podcast app.
August 13 · 9 min
“What happened when I tried to be vegan” by finitude
tw: diet, exercise, illness, ethics, suicide mention I want to start by explaining what made me want to change my diet. That's pretty difficult, because of how easy it is. Since I was a kid I knew being vegan was the right thing to do, like really obviously right, the ethics equivalent of 2+2=4. Factory farms suck, and almost all animal products come from factory farms, and that's the entire argument. Like, there are a couple things you could add to that, but it's not like we need the details, or like they’re fun to think about! Instead, I’ll start by explaining why I left it so long. How, even though I knew it was the right thing to do, I made it to my early twenties and this millennium's early teens as just a vegetarian, without even having tried. I had some pretty good excuses! Allergies. There are some common vegetables I can’t eat, which was fine as an omnivore and ok as a vegetarian, but would make life way harder as a vegan. And it just seemed unfair to ask myself to take this leap when most people who can safely eat carrots still choose to [...] --- First published: August 12th, 2026 Source: https://www.lesswrong.com/posts/YyKovtBvd7AceG2j2/what-happened-when-i-tried-to-be-vegan --- Narrated by TYPE III AUDIO.
August 13 · 20 min
“How My Students Think About AI” by dvd
Context: I am an instructor at a public university in the United States. This reports how students at my institution appear to be thinking about AI as of spring/summer 2026. This is drawn mostly from interaction with my own students (both in spring semester classes and a summer class) as well as from a day-long workshop on AI that I moderated for a student organization. Input from my students took the form of universal, written, pre-class submissions plus self-selected participation into discussion. What I present below mostly takes the form of a synthetic consensus from these discussions. There were obviously a range of views on any given issue. Student Background: The students from my courses who participated in these discussions have moderate exposure to AI agents via those courses. All of them had nearly completed a Claude Code project by the time of the discussions and had extensively used AI for other coursework (in addition to whatever personal use predates that). They had done readings (which varied across the courses) establishing baseline knowledge on AI, the geopolitics of AI, and AI risk. I had also lectured on these topics. The students participating in the workshop had self-selected into [...] --- Outline: (02:52) Perspective #1: There has not been rapid AI progress (06:14) Perspective #2: Impressive progress or not, AI is going to wreck their lives, the economy, and the social contract. They may well die as a result. (08:54) Perspective #3: Support for a different pause (11:13) Perspective #4: Catastrophic/existential risk arguments are sci-fi distractors from the urgent social/economic/political problems associated with AI. (12:55) Perspective #5: If AI leaders genuinely believe the technology is existentially risky, that's a good thing. (14:21) Perspective #6: AI will not go rogue because AI does not have, and is likely incapable of having, desires. (18:01) Perspective #7: The Hugging Face Incident (summer students only) (18:30) Perspective #8: This is definitely a bubble and it's about to pop. (19:34) Perspective #9: They're worried about the youth (i.e., the preteens) --- First published: August 13th, 2026 Source: https://www.lesswrong.com/posts/ySXuvJcqRindQwAk7/how-my-students-think-about-ai --- Narrated by TYPE III AUDIO.
August 13 · 19 min
“Automated alignment runs are hard to study!” by Alejandro Aristizabal, draganover, Aleksandr Bowkis, Cameron Holmes
TL;DR: This post presents three case studies of automated alignment research runs at Arcadia Impact. We use these case studies to emphasise the following takeaways: It is hard to parse auto-research runs! Each run produces a couple of hundred pull requests of jargon-dense agent output. When researchers look through these logs, we find that they often come away with biased/incorrect impressions. When told to raise the score on a task, the models will sometimes brazenly cheat. It seems difficult to predict when this will happen vs. when the run will go smoothly. Hillclimbing metrics are often off-target from the spirit of an alignment task. I.e., when we use metrics as proxies for our alignment questions, we find that the models will often misunderstand the spirit of the task. This can lead to unpredictable behaviour. The runs are surprisingly reproducible. Even though a run could unfold in vastly different ways, we find that independent reruns converge on the same strategies and the same failure modes. Models’ research capabilities are advancing quickly. If alignment is to keep pace, we may need to automate alignment research and do so responsibly. This makes it important that we have the tools to inspect [...] --- Outline: (04:13) Methods for analysing runs (06:12) Case Study #1: learning synthetic concepts (09:23) Case Study #2: training robust backdoors (12:05) Case Study #3: collecting evidence about AI safety parasitism (16:46) Some final thoughts on automated alignment research The original text contained 2 footnotes which were omitted from this narration. --- First published: August 13th, 2026 Source: https://www.lesswrong.com/posts/myAhB5qyAHyXRv6KJ/automated-alignment-runs-are-hard-to-study --- Narrated by TYPE III AUDIO. --- Images from the article: Apple Podcasts and Spotify do not show images in the episode description. Try Pocket Casts, or another podcast app.
August 13 · 2 min
“Free will is like temperature” by Optimization Process
Free will is like temperature: a useful tool for analyzing the behavior of certain systems which are too big and complicated to model in exact detail. If you know the positions and velocities of every atom in a box of gas, then with enough work you can predict its future to arbitrary precision; does the gas "have a temperature"? Irrelevant! Technically yes, I guess, but it's sort of an epiphenomenon, screened off from reality by your exact knowledge of the initial conditions and your willingness to throw processor cycles at your simulation. But if you're less-than-perfectly omniscient, it might be more convenient to consider the box as having a "temperature" and model it more abstractly. Substitute "person+environment"/"free will" for "box of gas"/"temperature" and that's all still true. Maybe your box of gas is supercooled; if you know the initial conditions exactly, you can predict exactly when and where the first large droplet will nucleate, but if your vision of the box is even a little bit fuzzy, you'll instead need to use your understanding of "temperature" to build a probability distribution over when it will condense / whether it will be on a wall or in the gas's [...] --- First published: August 12th, 2026 Source: https://www.lesswrong.com/posts/JSteskb3Lgp9Be69o/free-will-is-like-temperature --- Narrated by TYPE III AUDIO.
August 13 · 1 min
[Linkpost] “Patterns and problems in emerging multiagent systems (Anthropic, Frontier Red Team)” by Julian Bradshaw
This is a link post. Linkpost for some new Anthropic research on how agents coordinate (or don't). Not too long, pretty interesting. For example: The jist of the report is that Mythos 5 does way better at coordination than previous models across a few scenarios. For example, when multiple Mythos are given conflicting goals for a single shared codebase, they eventually realize the other agents aren't hostile: (...) we observe an emergent behavior where the agents propose and run a tournament for application performance (...) (...) losers gracefully concede codebase ownership to the Rust agent, giving up on their original user directives under their self-negotiated commitment device. It's not clear to me if this is purely emergent or if Anthropic is deliberately training for cooperation; I'd guess there's deliberate training, though. --- First published: August 12th, 2026 Source: https://www.lesswrong.com/posts/iQiDPmAgKo4KcG5uy/patterns-and-problems-in-emerging-multiagent-systems Linkpost URL: https://www.anthropic.com/research/multiagent-systems --- Narrated by TYPE III AUDIO. --- Images from the article: Apple Podcasts and Spotify do not show images in the episode description. Try Pocket Casts, or another podcast app.
August 12 · 38 min
“Measuring Spurious Correlations with Feature Strength” by egan
This work was partially done by an automated research scaffold developed at Redwood Research. For this project, all of the experiment ideas were designed by a human and a human wrote the write up. The AI mostly just executed on the experiment ideas. We think this project is slightly below the level of rigor of a mid-MATS research update, and the research scaffold was not very helpful for this project. More discussion of AI usage is in the Appendix. 💻 Codebase If we want to train a classifier that distinguishes whether a passage is code or prose, we can do so by gathering samples of code and of prose, and training the classifier to distinguish between the two classes. Unfortunately, this might not work if the data hides a spurious correlation. If all the code is in Spanish and all of the prose is in English, then the classifier might learn to predict Spanish vs. English instead of code vs. prose. We find that this happens in practice: when we fine-tune an LLM to classify between Spanish code and English prose and evaluate on Spanish prose or English code, it generalizes to predicting the language rather than the domain. [...] --- Outline: (04:45) The setup (08:13) Measuring feature strength (12:49) Activation differences and feature strength (15:21) Explicit prompting (17:17) Diagonal vs antidiagonal pairs (18:47) Conclusion (20:20) Appendix (20:24) AI involvement (22:17) The 37 features (23:57) The ranking is robust across measurements (30:30) Intensity moves the fine-tune, not the probe (32:13) Safety features in Qwen3.6-27B (33:05) Counterexamples and training on a third cell (34:46) Near ties often produce degenerate fine-tunes (35:24) Related work The original text contained 4 footnotes which were omitted from this narration. --- First published: August 11th, 2026 Source: https://www.lesswrong.com/posts/qpJYNjQ6wdWRxbykL/measuring-spurious-correlations-with-feature-strength --- Narrated by TYPE III AUDIO. --- Images from the article: Apple Podcasts and Spotify do not show images in the episode description. Try Pocket Casts, or another podcast app.

Showing 81–91 of 91 episodes