Claude Opus 4.6 is here. It was built with and mostly evaluated by Claude.

Their headline pitch includes:

Other notes:

---

Outline:

(03:45) A Three Act Play

(04:57) Safety Not Guaranteed

(10:53) Pliny Can Still Jailbreak Everything

(12:48) Transparency Is Good: The 212-Page System Card

(13:53) Mostly Harmless

(17:45) Mostly Honest

(19:01) Agentic Safety

(20:27) Prompt Injection

(23:07) Key Alignment Findings

(33:48) Behavioral Evidence (6.2)

(38:40) Reward Hacking and 'Overly Agentic Actions'

(40:37) Metrics (6.2.5.2)

(42:40) All I Did It All For The GUI

(43:58) Case Studies and Targeted Evaluations Of Behaviors (6.3)

(44:19) Misrepresenting Tool Results

(45:09) Unexpected Language Switching

(46:12) The Ghost of Jones Foods

(47:54) Loss of Style Points

(48:54) White Box Model Diffing

(49:13) Model Welfare

---

First published:

Source:

---

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.

Audio note: this article contains 55 uses of latex notation, so the narration may be difficult to follow. There's a link to the original text in the episode description.

Prerequisite: basic familiarity with what a prediction market is

So you want to run a prediction market. You need a way for people to trade shares. What are your options?

CLOBs and Markets

If you were making a prediction market from scratch, you'd probably come up with a Central Limit Order Book (CLOB). Traders post BUY and SELL orders, stating what they're willing to buy and sell, and at what price, and you record these orders in your book.

• Alice posts: "I'll buy 20 YES shares at $0.60 each"
• Bob posts: "I'll sell 25 YES shares at $0.65 each"
• Carol posts: "I'll buy 10 YES shares at $0.61 each" When someone wants to trade, and doesn't want to wait for someone to fulfill their order, you match them with the best available offer. Pretty intuitive.

This system shows up directly in Hypixel Skyblock and other MMOs. The Bazaar lets you post orders and wait, or instantly fulfill existing orders. Have some Enchanted Iron to sell? You can [...]

---

Outline:

(00:34) CLOBs and Markets

(01:47) Enter the Market Maker

(02:55) Pricing Shares

(04:21) Creating New Shares

(05:29) A Natural Probability Formula

(06:28) Discrete Shares

(08:27) The Calculus of Market Making

(12:00) Constant Product Market Maker

(14:49) Conclusion

The original text contained 2 footnotes which were omitted from this narration.

---

First published:

Source:

---

Narrated by TYPE III AUDIO.

I've let a few LLMs take David Bourget's and David Chalmers' 2020 PhilPapers Survey and made a little dashboard you can use to navigate the data: https://www.lordscottish.com/philsurvey.html

You can select multiple models to pool their data and compare it against the views of philosophers.

Some notable things:

1. LLMs are much more into one-boxing, and the strongest models (Claude opus 4.6 & ChatGPT5.2) 100% one-box.

2. Gpt4o and Grok3 are least interested in immortality. Gpt4o reject immortality in 4 out of 5 runs.

3. LLMs are less into moral generalism than philosophers (which might make sense, given how we train them on lots of special cases without clear underlying ethical principle?).

4. All llms accept a priori knowledge (100% across all runs!). Which might again make sense given their (arguable) lack of sense data?

5. LLMs lean towards there being little philosophical knowledge (philosophers lean towards a lot).

6. LLMs want to revise race categories much more than philosophers (83% vs 32%). Unlike philosophers, they treat race and gender similarly.

7. LLMs are really into deflationary realism as their metaontology compared to philosophers (83% vs 28%). BUT they also accept that there is a hard problem of consciousness [...]

---

First published:

Source:

---

Narrated by TYPE III AUDIO.

Save the date: LessOnline will be back again in 2026! As usual, it will take place at Lighthaven in Berkeley, CA.

Further details will be posted to LessWrong, or subscribe here for further updates, such as ticket sales going live. Details about who's attending, housing, relation to Manifest will be forthcoming.

What is LessOnline?

A Festival of Writers Who are Wrong on the Internet[1]

LessOnline is a festival celebrating truth-seeking, optimization, and blogging. It's an opportunity to meet people you've only ever known by their LessWrong username or Substack handle... or hang out with them irl again if you met them at the first or second LessOnline!

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

---

First published:

Source:

Linkpost URL:
http://Less.Online/

---

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.

Coverage of Claude Opus 4.6 started yesterday with the mundane alignment and model welfare sections of the model card.

Today covers the kinds of safety I think matter most: Sabotage, deception, situational awareness, outside red teaming and most importantly the frontier, catastrophic and existential risks. I think it was correct to release Opus 4.6 as an ASL-3 model, but the process Anthropic uses is breaking down, and it not on track to reliably get the right answer on Opus 5.

Tomorrow I’ll cover benchmarks, reactions and the holistic takeaways and practical implications. I’m still taking it all in, but it seems clear to me that Claude Opus 4.6 is the best model out there and should be your daily driver, with or without Claude Code, on most non-coding tasks, but it is not without its weaknesses, in particular in writing and falling into generating more ‘AI slop’ style prose than Claude Opus 4.5.

For coding tasks, I presume that Opus 4.6 with Claude Code is the play, especially with Agent Teams and fast mode available, and I’m using it myself, but Codex with GPT-5.3-Codex-Max is also a strong model and a viable alternative, and a fully [...]

---

Outline:

(01:32) Sabotage, Deception and Evaluation Integrity

(03:42) Sandbagging On Dangerous Capability Evaluations

(06:01) Situational Awareness

(07:33) Inhibiting Evaluation Awareness (6.5)

(09:06) Self-Preference

(10:24) UK AISI Testing

(11:40) Apollo Research Testing

(14:24) Responsible Scaling Policy Evaluations

(15:45) CBRN (mostly Biology)

(18:43) Autonomy

(26:40) Autonomy Benchmarks

(29:53) Cyber

(31:27) Ship It Anyway

(33:40) You Are Not Ready

---

First published:

Source:

---

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.

Note: this article required conversations with a lot of people. A (hopefully) exhaustive, randomized list of everyone whose thoughts contributed to the article: Lachlan Munroe (Head of Automation at DTU Biosustain), Max Hodak (CEO of Science, former founder of Transcriptic), D.J. Kleinbaum (CEO of Emerald Cloud Labs), Keoni Gandall (former founder of Trilobio), Cristian Ponce (CEO of Tetsuwan Scientific), Brontë Kolar (CEO of Zeon Systems), Jason Kelly (CEO of Ginkgo Bioworks), Jun Axup Penman (COO of E11 Bio), Nish Bhat (current VC, ex-Color cofounder), Amulya Garimella (MIT PhD student), Shelby Newsad (VC at Compound), Michelle Lee (CEO of Medra), Charles Yang (Fellow at Renaissance Philanthropy), Chase Armer (Columbia PhD student), Ben Ray (current founder, ex-Retro Biosciences automation engineer), and Jake Feala (startup creation at Flagship Pioneering).

1. Introduction
2. Heuristics for lab robotics
   1. There are box robots, and there are arm robots
   2. Most lab protocols can be automated, they just often aren’t worth automating
   3. You can improve lab robotics by improving the translation layer, the hardware layer, or the intelligence layer
   4. All roads lead to Transcriptic
3. Conclusion

Introduction

I have never worked in a wet lab. The closest I’ve come to it was during my first semester of undergrad [...]

---

Outline:

(01:56) Introduction

(04:33) Heuristics for lab robotics

(04:37) There are box robots, and there are arm robots

(12:40) Most lab protocols can be automated, they just often arent worth automating

(19:30) You can improve lab robotics by improving the translation layer, the hardware layer, or the intelligence layer

(35:16) All roads lead to Transcriptic

(48:43) Conclusion

---

First published:

Source:

---

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.

Cross-posted from Telescopic Turnip

Recommended soundtrack for this post

As we all know, the march of technological progress is best summarized by this meme from Linkedin:

Inventors constantly come up with exciting new inventions, each of them with the potential to change everything forever. But only a fraction of these ever establish themselves as a persistent part of civilization, and the rest vanish from collective consciousness. Before shutting down forever, though, the alternate branches of the tech tree leave some faint traces behind: over-optimistic sci-fi stories, outdated educational cartoons, and, sometimes, some obscure accessories that briefly made it to mass production before being quietly discontinued.

The classical example of an abandoned timeline is the Glorious Atomic Future, as described in the 1957 Disney cartoon Our Friend the Atom. A scientist with a suspiciously German accent explains all the wonderful things nuclear power will bring to our lives:

Sadly, the glorious atomic future somewhat failed to materialize, and, by the early 1960s, the project to rip a second Panama canal by detonating a necklace of nuclear bombs was canceled, because we are ruled by bureaucrats who hate fun and efficiency.

While the Our-Friend-the-Atom timeline remains out of reach from most [...]

---

Outline:

(02:08) Microwave Cooking, for One

(04:59) Out of the frying pan, into the magnetron

(09:12) Tradwife futurism

(11:52) Youll microwave steak and pasta, and youll be happy

(17:17) Microvibes

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

---

First published:

Source:

---

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.

In November 2025, while working as a writing coach at the experimental 2025 Inkhaven writing/blogging residency at the Lighthaven conference center, I was interviewed by Adam Mastroianni of Experimental History on where I get my ideas (of interest to Inkhaven participants, who were required to write >500 words⧸day) and develop my writings. It can be seen as a followup to my Dwarkesh Patel interview.

Topics include:

How recent poems emerged from LLM collaboration and happy accidents; incubation periods and waiting for ideas to “snap into place”; pathological curiosity and cross-domain reading; my Wikipedia apprenticeship; self-experimentation as easy science and statistics practice; the pipeline model of why small advantages compound into large output differences; mental tools like Feynman's algorithm and modus tollens reversal; “blog brain” and pattern-matching experiences to essays; the two directions of essay-writing (universal → concrete versus concrete → universal); and my preference for wikis over blogs, and treating publishing as just another edit.

Below is a lightly edited transcript of my interview and audience Q&A (to condense, clarify, and add references).

---

Outline:

(03:13) Opening Speech

(05:34) Poems & Incubation

(11:54) Polymath

(13:52) The Apprenticeship

(16:49) Self-Experimentation

(20:46) The Writing Pipeline

(23:17) Tools For Thought

(28:10) Blog Brain: Thats A Post

(32:53) Essay Archetype: Universal ↔︎ Concrete

(37:01) The Voice: Ideas As Earworms

(38:37) Audience Q&A

(38:41) Modalities & Comparative Advantage

(41:49) Publishing Thresholds

(44:10) Wikis Vs Blogs

(51:00) LLM Followup Questions

The original text contained 4 footnotes which were omitted from this narration.

---

First published:

Source:

---

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.

Much of my thinking over the last year has focusing on understanding the concept of "distributed agents", as opposed to the "centralized agents" that the existing paradigm of rational agency describes. Roughly speaking, centralized agents are more efficient (as sometimes formalized by the notion of "coherence"), while distributed agents are more robust.

Unfortunately the latter property is hard to formalize, since it requires that you perform well even in unpredicted (and sometimes unpredictable) situations. I give some tentative characterizations of distributed agents below, but there's still a lot of work to be done to formally define distributed agents. And ultimately I'd like to go further, to understand how agents can have both properties—which is roughly what I mean by "coalitional agency".

I gave a talk on the distinction about seven months ago. I'd been hoping to write up the main ideas at more length, but since that doesn't look like it'll happen any time soon, I'm sharing the slides below. Hopefully they're reasonably comprehensible by themselves, but feel free to ask questions about any parts that are unclear.

See more on my interpretation of Yudkowsky here; note that he disagrees with my emphasis on compression though (as per the [...]

---

First published:

Source:

---

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.

Previously: Donations, The Third Year / Donations, The First Year

In 2025, like in all previous years, I did what I was supposed to do. As each paycheck came in, before I did anything else, I dutifully put ten percent of it away in my "donations" savings account, to be disbursed at the end of the year.

It is still there, burning a hole in my pocket. I am very confused, and very sad.

EA was supposed to be easy, especially if you're one of the old school Peter Singer-inflected ones giving largely to global health and poverty reduction. You just give the funds away to whatever GiveWell recommends.

But one big thing that came into focus for me last year was that there are large institutional players who make up the shortfall whenever those charities don't fundraise enough from their own donor bases.

It is wonderful that this happens, to be clear. It is wonderful that the charities doing very important work get to have more stable financial projections, year over year. But as an individual small donor, the feeling I get now is that I am not actually giving to the Against Malaria Foundation. Instead, I am [...]

---

First published:

Source:

---

Narrated by TYPE III AUDIO.