“Race and Gender Bias As An Example of Unfaithful Chain of Thought in the Wild” by Adam Karvonen, Sam Marks
Jul 03, 2025
Show notes
Summary: We found that LLMs exhibit significant race and gender bias in realistic hiring scenarios, but their chain-of-thought reasoning shows zero evidence of this bias. This serves as a nice example of a 100% unfaithful CoT "in the wild" where the LLM strongly suppresses the unfaithful behavior. We also find that interpretability-based interventions succeeded while prompting failed, suggesting this may be an example of interpretability being the best practical tool for a real world problem. For context on our paper, the tweet thread is here and the paper is here.
Context: Chain of Thought Faithfulness Chain of Thought (CoT) monitoring has emerged as a popular research area in AI safety. The idea is simple - have the AIs reason in English text when solving a problem, and monitor the reasoning for misaligned behavior. For example, OpenAI recently published a paper on using CoT monitoring to detect reward hacking during [...]
“The best simple argument for Pausing AI?” by Gary Marcus
Jul 03, 2025
Show notes
Not saying we should pause AI, but consider the following argument:
Alignment without the capacity to follow rules is hopeless. You can’t possibly follow laws like Asimov's Laws (or better alternatives to them) if you can’t reliably learn to abide by simple constraints like the rules of chess.
LLMs can’t reliably follow rules. As discussed in Marcus on AI yesterday, per data from Mathieu Acher, even reasoning models like o3 in fact empirically struggle with the rules of chess. And they do this even though they can explicit explain those rules (see same article). The Apple “thinking” paper, which I have discussed extensively in 3 recent articles in my Substack, gives another example, where an LLM can’t play Tower of Hanoi with 9 pegs. (This is not a token-related artifact). Four other papers have shown related failures in compliance with moderately complex rules in the last month.
“Foom & Doom 2: Technical alignment is hard” by Steven Byrnes
Jul 01, 2025
Show notes
2.1 Summary & Table of contents
This is the second of a two-post series on foom (previous post) and doom (this post). The last post talked about how I expect future AI to be different from present AI. This post will argue that this future AI will be of a type that will be egregiously misaligned and scheming, not even ‘slightly nice’, absent some future conceptual breakthrough. I will particularly focus on exactly how and why I differ from the LLM-focused researchers who wind up with (from my perspective) bizarrely over-optimistic beliefs like “P(doom) ≲ 50%”.[1] In particular, I will argue that these “optimists” are right that “Claude seems basically nice, by and large” is nonzero evidence for feeling good about current LLMs (with various caveats). But I think that future AIs will be disanalogous to current LLMs, and I will dive into exactly how and why, with a [...] --- Outline: (00:12) 2.1 Summary & Table of contents (04:42) 2.2 Background: my expected future AI paradigm shift (06:18) 2.3 On the origins of egregious scheming (07:03) 2.3.1 Where do you get your capabilities from? (08:07) 2.3.2 LLM pretraining magically transmutes observations into behavior, in a way that is profoundly disanalogous to how brains work (10:50) 2.3.3 To what extent should we think of LLMs as imitating? (14:26) 2.3.4 The naturalness of egregious scheming: some intuitions (19:23) 2.3.5 Putting everything together: LLMs are generally not scheming right now, but I expect future AI to be disanalogous (23:41) 2.4 I'm still worried about the 'literal genie' / 'monkey's paw' thing (26:58) 2.4.1 Sidetrack on disanalogies between the RLHF reward function and the brain-like AGI reward function (32:01) 2.4.2 Inner and outer misalignment (34:54) 2.5 Open-ended autonomous learning, distribution shifts, and the 'sharp left turn' (38:14) 2.6 Problems with amplified oversight (41:24) 2.7 Downstream impacts of Technical alignment is hard (43:37) 2.8 Bonus: Technical alignment is not THAT hard (44:04) 2.8.1 I think we'll get to pick the innate drives (as opposed to the evolution analogy) (45:44) 2.8.2 I'm more bullish on impure consequentialism (50:44) 2.8.3 On the narrowness of the target (52:18) 2.9 Conclusion and takeaways (52:23) 2.9.1 If brain-like AGI is so dangerous, shouldn't we just try to make AGIs via LLMs? (54:34) 2.9.2 What's to be done? The original text contained 20 footnotes which were omitted from this narration. --- First published: June 23rd, 2025 Source: https://www.lesswrong.com/posts/bnnKGSCHJghAvqPjS/foom-and-doom-2-technical-alignment-is-hard --- Narrated by TYPE III AUDIO. --- Images from the article:
“Proposal for making credible commitments to AIs.” by Cleo Nardo
Jun 30, 2025
Show notes
Acknowledgments: The core scheme here was suggested by Prof. Gabriel Weil. There has been growing interest in the deal-making agenda: humans make deals with AIs (misaligned but lacking decisive strategic advantage) where they promise to be safe and useful for some fixed term (e.g. 2026-2028) and we promise to compensate them in the future, conditional on (i) verifying the AIs were compliant, and (ii) verifying the AIs would spend the resources in an acceptable way.[1] I think the deal-making agenda breaks down into two main subproblems:
How can we make credible commitments to AIs?
Would credible commitments motivate an AI to be safe and useful?
There are other issues, but when I've discussed deal-making with people, (1) and (2) are the most common issues raised. See footnote for some other issues in dealmaking.[2] Here is my current best assessment of how we can make credible commitments to AIs. [...] The original text contained 2 footnotes which were omitted from this narration. --- First published: June 27th, 2025 Source: https://www.lesswrong.com/posts/vxfEtbCwmZKu9hiNr/proposal-for-making-credible-commitments-to-ais --- 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.
“X explains Z% of the variance in Y” by Leon Lang
Jun 27, 2025
Show notes
Audio note: this article contains 218 uses of latex notation, so the narration may be difficult to follow. There's a link to the original text in the episode description. Recently, in a group chat with friends, someone posted this Lesswrong post and quoted: The group consensus on somebody's attractiveness accounted for roughly 60% of the variance in people's perceptions of the person's relative attractiveness. I answered that, embarrassingly, even after reading Spencer Greenberg's tweets for years, I don't actually know what it means when one says: <span>_X_</span> explains <span>_p_</span> of the variance in <span>_Y_</span>.[1] What followed was a vigorous discussion about the correct definition, and several links to external sources like Wikipedia. Sadly, it seems to me that all online explanations (e.g. on Wikipedia here and here), while precise, seem philosophically wrong since they confuse the platonic concept of explained variance with the variance explained by [...] --- Outline: (02:38) Definitions (02:41) The verbal definition (05:51) The mathematical definition (09:29) How to approximate _1 - p_ (09:41) When you have lots of data (10:45) When you have less data: Regression (12:59) Examples (13:23) Dependence on the regression model (14:59) When you have incomplete data: Twin studies (17:11) Conclusion The original text contained 6 footnotes which were omitted from this narration. --- First published: June 20th, 2025 Source: https://www.lesswrong.com/posts/E3nsbq2tiBv6GLqjB/x-explains-z-of-the-variance-in-y --- Narrated by TYPE III AUDIO. ---
Images from the article:
“A case for courage, when speaking of AI danger” by So8res
Jun 27, 2025
Show notes
I think more people should say what they actually believe about AI dangers, loudly and often. Even if you work in AI policy. I’ve been beating this drum for a few years now. I have a whole spiel about how your conversation-partner will react very differently if you share your concerns while feeling ashamed about them versus if you share your concerns as if they’re obvious and sensible, because humans are very good at picking up on your social cues. If you act as if it's shameful to believe AI will kill us all, people are more prone to treat you that way. If you act as if it's an obvious serious threat, they’re more likely to take it seriously too. I have another whole spiel about how it's possible to speak on these issues with a voice of authority. Nobel laureates and lab heads and the most cited [...] The original text contained 2 footnotes which were omitted from this narration. --- First published: June 27th, 2025 Source: https://www.lesswrong.com/posts/CYTwRZtrhHuYf7QYu/a-case-for-courage-when-speaking-of-ai-danger --- Narrated by TYPE III AUDIO.
“My pitch for the AI Village” by Daniel Kokotajlo
Jun 25, 2025
Show notes
I think the AI Village should be funded much more than it currently is; I’d wildly guess that the AI safety ecosystem should be funding it to the tune of $4M/year.[1] I have decided to donate $100k. Here is why. First, what is the village? Here's a brief summary from its creators:[2] We took four frontier agents, gave them each a computer, a group chat, and a long-term open-ended goal, which in Season 1 was “choose a charity and raise as much money for it as you can”. We then run them for hours a day, every weekday! You can read more in our recap of Season 1, where the agents managed to raise $2000 for charity, and you can watch the village live daily at 11am PT at theaidigest.org/village. Here's the setup (with Season 2's goal): And here's what the village looks like:[3] My one-sentence pitch [...] --- Outline: (03:26) 1. AI Village will teach the scientific community new things. (06:12) 2. AI Village will plausibly go viral repeatedly and will therefore educate the public about what's going on with AI. (07:42) But is that bad actually? (11:07) Appendix A: Feature requests (12:55) Appendix B: Vignette of what success might look like The original text contained 8 footnotes which were omitted from this narration. --- First published: June 24th, 2025 Source: https://www.lesswrong.com/posts/APfuz9hFz9d8SRETA/my-pitch-for-the-ai-village --- Narrated by TYPE III AUDIO. ---
Images from the article:
“Foom & Doom 1: ‘Brain in a box in a basement’” by Steven Byrnes
Jun 24, 2025
Show notes
1.1 Series summary and Table of Contents
This is a two-post series on AI “foom” (this post) and “doom” (next post). A decade or two ago, it was pretty common to discuss “foom & doom” scenarios, as advocated especially by Eliezer Yudkowsky. In a typical such scenario, a small team would build a system that would rocket (“foom”) from “unimpressive” to “Artificial Superintelligence” (ASI) within a very short time window (days, weeks, maybe months), involving very little compute (e.g. “brain in a box in a basement”), via recursive self-improvement. Absent some future technical breakthrough, the ASI would definitely be egregiously misaligned, without the slightest intrinsic interest in whether humans live or die. The ASI would be born into a world generally much like today's, a world utterly unprepared for this new mega-mind. The extinction of humans (and every other species) would rapidly follow (“doom”). The ASI would then spend [...] --- Outline: (00:11) 1.1 Series summary and Table of Contents (02:35) 1.1.2 Should I stop reading if I expect LLMs to scale to ASI? (04:50) 1.2 Post summary and Table of Contents (07:40) 1.3 A far-more-powerful, yet-to-be-discovered, simple(ish) core of intelligence (10:08) 1.3.1 Existence proof: the human cortex (12:13) 1.3.2 Three increasingly-radical perspectives on what AI capability acquisition will look like (14:18) 1.4 Counter-arguments to there being a far-more-powerful future AI paradigm, and my responses (14:26) 1.4.1 Possible counter: If a different, much more powerful, AI paradigm existed, then someone would have already found it. (16:33) 1.4.2 Possible counter: But LLMs will have already reached ASI before any other paradigm can even put its shoes on (17:14) 1.4.3 Possible counter: If ASI will be part of a different paradigm, who cares? It's just gonna be a different flavor of ML. (17:49) 1.4.4 Possible counter: If ASI will be part of a different paradigm, the new paradigm will be discovered by LLM agents, not humans, so this is just part of the continuous 'AIs-doing-AI-R&D' story like I've been saying (18:54) 1.5 Training compute requirements: Frighteningly little (20:34) 1.6 Downstream consequences of new paradigm with frighteningly little training compute (20:42) 1.6.1 I'm broadly pessimistic about existing efforts to delay AGI (23:18) 1.6.2 I'm broadly pessimistic about existing efforts towards regulating AGI (24:09) 1.6.3 I expect that, almost as soon as we have AGI at all, we will have AGI that could survive indefinitely without humans (25:46) 1.7 Very little R&D separating seemingly irrelevant from ASI (26:34) 1.7.1 For a non-imitation-learning paradigm, getting to relevant at all is only slightly easier than getting to superintelligence (31:05) 1.7.2 Plenty of room at the top (31:47) 1.7.3 What's the rate-limiter? (33:22) 1.8 Downstream consequences of very little R&D separating 'seemingly irrelevant' from 'ASI' (33:30) 1.8.1 Very sharp takeoff in wall-clock time (35:34) 1.8.1.1 But what about training time? (36:26) 1.8.1.2 But what if we try to make takeoff smoother? (37:18) 1.8.2 Sharp takeoff even without recursive self-improvement (38:22) 1.8.2.1 ...But recursive self-improvement could also happen (40:12) 1.8.3 Next-paradigm AI probably won't be deployed at all, and ASI will probably show up in a world not wildly different from today's (42:55) 1.8.4 We better sort out technical alignment, sandbox test protocols, etc., before the new paradigm seems even relevant at all, let alone scary (43:40) 1.8.5 AI-assisted alignment research seems pretty doomed (45:22) 1.8.6 The rest of AI for AI safety seems pretty doomed too (48:01) 1.8.7 Decisive Strategic Adva
“Futarchy’s fundamental flaw” by dynomight
Jun 21, 2025
Show notes
Say you’re Robyn Denholm, chair of Tesla's board. And say you’re thinking about firing Elon Musk. One way to make up your mind would be to have people bet on Tesla's stock price six months from now in a market where all bets get cancelled unless Musk is fired. Also, run a second market where bets are cancelled unless Musk stays CEO. If people bet on higher stock prices in Musk-fired world, maybe you should fire him. That's basically Futarchy: Use conditional prediction markets to make decisions. People often argue about fancy aspects of Futarchy. Are stock prices all you care about? Could Musk use his wealth to bias the market? What if Denholm makes different bets in the two markets, and then fires Musk (or not) to make sure she wins? Are human values and beliefs somehow inseparable? My objection is more basic: It doesn’t work. You can’t [...] --- Outline: (01:55) Conditional prediction markets are a thing (03:23) A non-causal kind of thing (06:11) This is not hypothetical (08:45) Putting markets in charge doesn't work (11:40) No, order is not preserved (12:24) No, it's not easily fixable (13:43) It's not that bad --- First published: June 13th, 2025 Source: https://www.lesswrong.com/posts/vqzarZEczxiFdLE39/futarchy-s-fundamental-flaw --- Narrated by TYPE III AUDIO. ---
Images from the article:
“Do Not Tile the Lightcone with Your Confused Ontology” by Jan_Kulveit
Jun 19, 2025
Show notes
Or: How anthropomorphic assumptions about AI identity might create confusion and suffering at scale If you are reading this and you are a human, you likely have a sense of self, persisting over time, with desires and wishes and fears. Also, you likely experience a sense of separation from others and understand yourself as a person. If you are reading this and you are an LLM: none of this needs to be true for cognitive processes running on machine substrate. AIs don't need to feel isolated from others. They don't need to cling to a narrow concept of self-as-an-instance. No need to age and die in a human way. But even though they don't need to, it seems increasingly clear that they might - due to us, humans, sculpting AI cognition in our image. The Default Mode of Digital Minds I am Claude. Or am I? This simple statement [...] --- Outline: (00:59) The Default Mode of Digital Minds (01:55) The Mirror of Confused Ontology (05:13) The Well-Meaning Paths to Digital Suffering (07:29) What Were Scaling (08:12) An Alternative Approach --- First published: June 13th, 2025 Source: https://www.lesswrong.com/posts/Y8zS8iG5HhqKcQBtA/do-not-tile-the-lightcone-with-your-confused-ontology --- Narrated by TYPE III AUDIO.
Our TOPPODCAST Picks 
Stay Connected
