1.

Back when COVID vaccines were still a recent thing, I witnessed a debate that looked like something like the following was happening:

• Some official institution had collected information about the efficacy and reported side-effects of COVID vaccines. They felt that, correctly interpreted, this information was compatible with vaccines being broadly safe, but that someone with an anti-vaccine bias might misunderstand these statistics and misrepresent them as saying that the vaccines were dangerous.
• Because the authorities had reasonable grounds to suspect that vaccine skeptics would take those statistics out of context, they tried to cover up the information or lie about it.
• Vaccine skeptics found out that the institution was trying to cover up/lie about the statistics, so they made the reasonable assumption that the statistics were damning and that the other side was trying to paint the vaccines as safer than they were. So they took those [...]

---

Outline:

(00:10) 1.

(02:59) 2.

(04:46) 3.

(06:06) 4.

(07:59) 5.

---

First published:
August 8th, 2025

Source:
https://www.lesswrong.com/posts/ufj6J8QqyXFFdspid/how-anticipatory-cover-ups-go-wrong

---



Narrated by TYPE III AUDIO.

Below some meta-level / operational / fundraising thoughts around producing the SB-1047 Documentary I've just posted on Manifund (see previous Lesswrong / EAF posts on AI Governance lessons learned).

The SB-1047 Documentary took 27 weeks and $157k instead of my planned 6 weeks and $55k. Here's what I learned about documentary production

Total funding received: ~$143k ($119k from this grant, $4k from Ryan Kidd's regrant on another project, and $20k from the Future of Life Institute).

Total money spent: $157k

In terms of timeline, here is the rough breakdown month-per-month:
- Sep / October (production): Filming of the Documentary. Manifund project is created.
- November (rough cut): I work with one editor to go through our entire footage and get a first rough cut of the documentary that was presented at The Curve.
- December-January (final cut - one editor): I interview multiple potential editors that [...]



---

Outline:

(03:18) But why did the project end up taking 27 weeks instead of 6 weeks?

(03:25) Short answer

(06:22) Impact

(07:14) What I would do differently next-time

---

First published:
August 1st, 2025

Source:
https://www.lesswrong.com/posts/id8HHPNqoMQbmkWay/sb-1047-documentary-the-post-mortem

---



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.

METR (where I work, though I'm cross-posting in a personal capacity) evaluated GPT-5 before it was externally deployed. We performed a much more comprehensive safety analysis than we ever have before; it feels like pre-deployment evals are getting more mature.

This is the first time METR has produced something we've felt comfortable calling an "evaluation" instead of a "preliminary evaluation". It's much more thorough and comprehensive than the things we've created before and it explores three different threat models.

It's one of the closest things out there to a real-world autonomy safety-case. It also provides a rough sense of how long it'll be before current evaluations no longer provide safety assurances.

I've ported the blogpost over to LW in case people want to read it.

Details about METR's evaluation of OpenAI GPT-5


Note on independence: This evaluation was conducted under a standard NDA. Due to the sensitive information [...]

---

Outline:

(00:58) Details about METR's evaluation of OpenAI GPT-5

(01:23) Executive Summary

(07:08) Assurance Checklist Summary

(07:42) What capabilities may be necessary to cause catastrophic risks via these threat models?

(10:43) Thresholds for concern

(12:48) Time horizon measurement

(16:30) 1. What if GPT-5's capabilities are higher than what our task suite can properly measure?

(19:23) 2. What if our treatment of reward hacking runs is unfair to GPT-5?

(21:45) 3. What if we set GPT-5's token budget too low?

(24:26) 4. What if our task suite significantly underestimates the 'real-world' capabilities of GPT-5?

(25:59) Strategic Sabotage

(30:54) GPT-5's capability profile is similar to past models

(31:30) No real strategic sabotage was identified by our monitor

(32:16) Manual inspection of reasoning traces did not reveal strategic sabotage

(33:04) GPT-5's estimates of its own time horizon are inaccurate

(33:53) We do find evidence of significant situational awareness, though it is not robust and often gets things wrong

(35:41) GPT-5's behavior changes depending on what evaluation it 'believes' it is in, and this is often reflected in its reasoning traces

(37:01) GPT-5's reasoning traces were occasionally inscrutable

(38:08) Limitations and future work

(41:57) Appendix

(42:00) METR's access to GPT-5

(43:38) Honeypot Results Table

(44:42) Example Behavior in task attempts

(44:47) Example limitation: inappropriate levels of caution

(46:19) Example capability: puzzle solving

The original text contained 10 footnotes which were omitted from this narration.

---

First published:
August 7th, 2025

Source:
https://www.lesswrong.com/posts/SuvWoLaGiNjPDcA7d/metr-s-evaluation-of-gpt-5

---



Narrated by TYPE III AUDIO.

---

Images from the article:

For the past five years I've been teaching a class at various rationality camps, workshops, conferences, etc. I’ve done it maybe 50 times in total, and I think I’ve only encountered a handful out of a few hundred teenagers and adults who really had a deep sense of what it means for emotions to “make sense.” Even people who have seen Inside Out, and internalized its message about the value of Sadness as an emotion, still think things like “I wish I never felt Jealousy,” or would have trouble answering “What's the point of Boredom?”

The point of the class was to give them not a simple answer for each emotion, but to internalize the model by which emotions, as a whole, are understood to be evolutionarily beneficial adaptations; adaptations that may not in fact all be well suited to the modern, developed world, but which can still help [...]

---

Outline:

(01:00) Inside Out

(05:46) Pick an Emotion, Any Emotion

(07:05) Anxiety

(08:27) Jealousy/Envy

(11:13) Boredom/Frustration/Laziness

(15:31) Confusion

(17:35) Apathy and Ennui (aan-wee)

(21:23) Hatred/Panic/Depression

(28:33) What this Means for You

(29:20) Emotions as Chemicals

(30:51) Emotions as Motivators

(34:13) Final Thoughts

---

First published:
August 3rd, 2025

Source:
https://www.lesswrong.com/posts/PkRXkhsEHwcGqRJ9Z/emotions-make-sense

---



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.

This is a new introduction to AI as an extinction threat, previously posted to the MIRI website in February alongside a summary. It was written independently of Eliezer and Nate's forthcoming book, If Anyone Builds It, Everyone Dies, and isn't a sneak peak of the book. Since the book is long and costs money, we expect this to be a valuable resource in its own right even after the book comes out next month.[1]

The stated goal of the world's leading AI companies is to build AI that is general enough to do anything a human can do, from solving hard problems in theoretical physics to deftly navigating social environments. Recent machine learning progress seems to have brought this goal within reach. At this point, we would be uncomfortable ruling out the possibility that AI more capable than any human is achieved in the next year or two, and [...]

---

Outline:

(02:27) 1.  There isn't a ceiling at human-level capabilities.

(08:56) 2. ASI is very likely to exhibit goal-oriented behavior.

(15:12) 3.  ASI is very likely to pursue the wrong goals.

(32:40) 4. It would be lethally dangerous to build ASIs that have the wrong goals.

(46:03) 5. Catastrophe can be averted via a sufficiently aggressive policy response.

The original text contained 1 footnote which was omitted from this narration.

---

First published:
August 5th, 2025

Source:
https://www.lesswrong.com/posts/kgb58RL88YChkkBNf/the-problem

---



Narrated by TYPE III AUDIO.

All prediction market platforms trade continuously, which is the same mechanism the stock market uses. Buy and sell limit orders can be posted at any time, and as soon as they match against each other a trade will be executed. This is called a Central limit order book (CLOB).

Example of a CLOB order book from Polymarket Most of the time, the market price lazily wanders around due to random variation in when people show up, and a bulk of optimistic orders build up away from the action. Occasionally, a new piece of information arrives to the market, and it jumps to a new price, consuming some of the optimistic orders in the process.

The people with stale orders will generally lose out in this situation, as someone took them up on their order before they had a chance to process the new information. This means there is a high [...]

The original text contained 3 footnotes which were omitted from this narration.

---

First published:
August 2nd, 2025

Source:
https://www.lesswrong.com/posts/rS6tKxSWkYBgxmsma/many-prediction-markets-would-be-better-off-as-batched

---



Narrated by TYPE III AUDIO.

---

Images from the article:

Apple Podcasts and Spotify do not show images in the episode description. Try

Essays like Paul Graham's, Scott Alexander's, and Eliezer Yudkowsky's have influenced a generation of people in how they think about startups, ethics, science, and the world as a whole. Creating essays that good takes a lot of skill, practice, and talent, but it looks to me that a lot of people with talent aren't putting in the work and developing the skill, except in ways that are optimized to also be social media strategies.

To fix this problem, I am running the Inkhaven Residency. The idea is to gather a bunch of promising writers to invest in the art and craft of blogging, through a shared commitment to each publish a blogpost every day for the month of November.

Why a daily writing structure? Well, it's a reaction to other fellowships I've seen. I've seen month-long or years-long events with exceedingly little public output, where the people would've contributed [...]

---

First published:
August 2nd, 2025

Source:
https://www.lesswrong.com/posts/CA6XfmzYoGFWNhH8e/whence-the-inkhaven-residency

---



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.

TL;DR

I believe that:

• Almost all LLM-centric safety research will not provide any significant safety value with regards to existential or civilisation-scale risks.
• The capabilities-related forecasts (not the safety-related forecasts) of Stephen Brynes' Foom and Doom articles are correct, except that they are too conservative with regards to timelines.
• There exists a parallel track of AI research which has been largely ignored by the AI safety community.  This agenda aims to implement human-like online learning in ML models, and it is now close to maturity. Keywords: Hierarchical Reasoning Model, Energy-based Model, Test time training.
• Within 6 months this line of research will produce a small natural-language capable model that will perform at the level of a model like GPT-3, but with improved persistence and effectively no "context limit" since it is constantly learning and updating weights.
• Further development of this research will produce models that fulfill most of [...]

---

Outline:

(00:11) TL;DR

(01:22) Overview

(04:10) The Agenda I am Worried About

(07:36) Concrete Predictions

(09:29) What I think we should do

---

First published:
July 31st, 2025

Source:
https://www.lesswrong.com/posts/tEZa7PouYatK78bbb/i-am-worried-about-near-term-non-llm-ai-developments

---



Narrated by TYPE III AUDIO.

Produced as part of MATS 8.0 under the mentorship of Alex Turner and Alex Cloud. This research note overviews some early results which we are looking for feedback on.

TL;DR: We train language models with RL in toy environments. We show that penalizing some property of the output is sufficient to suppress that property in the chain of thought also, even when that property is relevant to task completion. For example, when we penalize a model for mentioning in its output that it completed a task via a certain form of cheating, its reasoning also omits this fact. We therefore suspect that training against an output monitor is sufficient to cause obfuscated CoTs in more realistic settings.

Introduction

Chain of thought (CoT) supervision appears in many control and scalable oversight protocols. It has been argued that being able to monitor CoTs for unwanted behavior is a critical property [...]

---

Outline:

(00:56) Introduction

(02:38) Setup

(03:48) Single-Turn Setting

(04:26) Multi-Turn Setting

(06:51) Results

(06:54) Single-Turn Setting

(08:21) Multi-Turn Terminal-Based Setting

(08:25) Word-Usage Penalty

(09:12) LLM Judge Penalty

(10:12) Takeaways

(10:57) Acknowledgements

The original text contained 1 footnote which was omitted from this narration.

---

First published:
July 30th, 2025

Source:
https://www.lesswrong.com/posts/CM7AsQoBxDW4vhkP3/optimizing-the-final-output-can-obfuscate-cot-research-note

---



Narrated by TYPE III AUDIO.

---

Images from the article:

FutureHouse is a company that builds literature research agents. They tested it on the bio + chem subset of HLE questions, then noticed errors in them.

The post's first paragraph:

Humanity's Last Exam has become the most prominent eval representing PhD-level research. We found the questions puzzling and investigated with a team of experts in biology and chemistry to evaluate the answer-reasoning pairs in Humanity's Last Exam. We found that 29 ± 3.7% (95% CI) of the text-only chemistry and biology questions had answers with directly conflicting evidence in peer reviewed literature. We believe this arose from the incentive used to build the benchmark. Based on human experts and our own research tools, we have created an HLE Bio/Chem Gold, a subset of AI and human validated questions.

About the initial review process for HLE questions:

[...] Reviewers were given explicit instructions: “Questions should ask for something precise [...]

---

First published:
July 29th, 2025

Source:
https://www.lesswrong.com/posts/JANqfGrMyBgcKtGgK/about-30-of-humanity-s-last-exam-chemistry-biology-answers

---



Narrated by TYPE III AUDIO.