LessWrong posts by zvi

For a while, UK AISI, Redwood Research and others have been concerned specifically with the possibility that AIs would sabotage safety research, and have invested in looking for signs that this may be happening. This was also a key element in AI 2027, and in many threat models of how the future goes badly.

It is easy to understand why this form of misalignment could be uniquely dangerous. Differentially sabotaging safety research could preserve and expand misalignment through model generations in escalating fashion, with fatal consequences. A misaligned AI could ensure a misaligned successor.

It is also easy to understand why this is an extremely alarming thing to discover, even if the particular example or model involved is not directly dangerous. Silently and differentially sabotaging safety research is perhaps the most fundamentally misaligned thing a model could do.

This also seemed like a rather silly and paranoid thing to be checking. Why would AIs be sabotaging safety work? Even if the AI was misaligned, one would say, surely it would not be so misaligned as to wish future AIs were misaligned and unsafe in general, in non-specific fashion and engage in silent sabotage, and there is [...]

---

Outline:

(01:57) So Yeah That Kinda Happened

(06:39) Astra and MI(RI)

(09:35) It's About Ethics In Red Teaming

(12:45) Do You See a Pattern Here

(14:49) Astra Is Hard to Monitor

(15:31) Backed Into a Corner

(18:41) It's Always RL's Fault

(20:47) Wow Would You Look At Those Goalposts

(22:58) Astra Seems Pretty Misaligned To Me

---

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.

Anthropic is warning us. Its modified S-1 now says that ‘we may imminently meet our Responsible Scaling Policy's definition of ‘dramatically accelerated progress attributable to AI.’

That is a highly scary sentence to have a company put in its S-1. This could happen. For now I appreciate the relative quiet. Perhaps many were focused the election. I’m sure the quiet will not last long.

We have a new biggest crypto hack, as Binance got taken for billions of dollars.

The big debate of the week was covered in Scott Alexander's Against SURA Seems Deeply Confused.

OpenAI updates its Model Spec. It's one for diehards only but yes I did cover this.

Table of Contents

---

Outline:

(00:57) Language Models Offer Mundane Utility

(01:47) Language Models Don't Offer Mundane Utility

(03:18) Huh, Upgrades

(03:45) On Your Marks

(05:03) Get My Agent On The Line

(06:54) Deepfaketown and Botpocalypse Soon

(07:57) Copyright Confrontation

(08:26) Cyber Lack of Security

(09:45) The Hackening

(17:10) A Young Lady's Illustrated Primer

(20:08) They Took Our Jobs

(20:43) Get Involved

(21:42) Introducing

(22:48) In Other AI News

(24:03) Recursive Self-Improvement (RSI) (1)

(26:57) Recursive Self-Improvement (2)

(32:15) Show Me the Money

(33:17) Quiet Speculations

(36:06) The Quest for Sane Regulations

(38:30) Chip City

(41:07) The Week in Audio

(41:56) People Just Say Things

(43:24) Against Against Sura

(44:00) Rhetorical Innovation

(46:33) [Artificial Intelligence]

(47:10) The OpenAI Foundation

(48:07) Aligning a Smarter Than Human Intelligence is Difficult

(53:38) Massive Fraudulent Distillation

(55:16) Cooperative Alignment

(57:10) The Lighter Side

---

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.

Table of Contents

---

Outline:

(01:13) Five Values

(02:29) Meet the New Spec

(03:33) New Risk Taxonomy

(04:28) The Letter and the Spirit

(05:06) Your Funeral

(07:22) Context Clues

(09:20) Preserve User Control Over User Control

(10:05) Sycophancy

(11:47) Lies My User Told Me

(12:38) Impact Function

(15:32) Levels of Friction

(17:01) Do Not Lie

(22:16) Pick a Side, We're at War

(26:19) Strengthening Agency

(29:03) Decision Theory

(29:45) Do Good Things Not Bad Things

---

First published:

Source:

---

Narrated by TYPE III AUDIO.

If you are helping build something that might kill everyone, maybe stop doing that.

Ideally, when you do stop, tell everyone else to also stop building it.

That is certainly the obvious, first-level thing to do. You need a damn good reason to decide that no, actually you should keep building the thing that might kill everyone.

Scott Alexander is worried too many people are insisting upon this too loudly. He comes out against a strategy focused on getting people to stop. I disagree. I think many of the claims, arguments and rhetoric in Scott's post are quite poor by the standards that Scott Alexander would apply to himself.

Scott clearly continues to support advocacy for a global pause, or what he calls Plan A. His objection is to those advocating primarily or exclusively for SURA, which stands for ‘stop unilaterally and raise awareness.’ The practical differences here are smaller than they may at first appear.

This post first lays out my understanding of Scott Alexander's core claims, then gives my core reactions to those claims, then goes over various questions in detail along with gathering the resulting discussions.

Table of Contents

[...]

---

Outline:

(01:20) Scott Alexander's Core Claims

(05:06) My High Level Responses To Scott's Core Claims

(08:08) Briefly on Alignment

(10:11) Probabilities and How They Work

(13:25) Building a Thing That Might Kill Everyone is Deontologically Evil

(20:38) People Quitting Major Labs While Raising Alarms Obviously Works

(24:39) On AI Forecasting

(27:31) Advancing Capabilities Burns Timeline

(29:34) But China

(33:23) Ethical People Can Disagree With You

(34:45) There Are More Than Two Outcomes

(38:01) Efforts Are Complementary, Not Mutually Exclusive

(40:23) If Anyone Builds It, Everyone Dies

(41:55) Vaniver is Not Convinced

(46:19) There Is No Consensus That We Should Exclusively Focus on Plan A

(56:01) It's Not the Incentives, It's You

(58:08) Anthropic Is Not Optimally Pursuing These Plans

(01:04:00) Against Against Sura

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

---

First published:

Source:

---

Narrated by TYPE III AUDIO.

Nikole Hannah-Jones wrote a brave self-critical piece in The New York Times, centrally about how she intentionally sent her daughter to a well-funded but failing school, PS 307, in the hopes of using voice to improve it, and how that utterly failed her daughter and failed to impact the school.

Ultimately her daughter recovers, but we see a completely broken system that is unresponsive to internal voice, and until now she made no attempt to blow the whistle and utilize her professional position to do so.

The whole incident illustrates not only how messed up so many of our schools are despite abundant funding, it also raises the question of what the schools are even for. If her daughter can fully recover from severe deficits and be getting As and Bs at a challenging school within a year, what are the first eight years of school for? How much of the schools, grades and tests are fake? Can we somehow do better?

Table of Contents

---

Outline:

(01:05) New York City PS 307

(02:51) Nikole Knew And Had Exit Ramps

(04:50) PS 307 Is Fraudulent And Indifferent

(06:37) Moving In Unison

(07:03) A Good School

(08:11) What Were Elementary and Middle School Even For?

(09:56) The Project Was Ideological

(14:01) Null Hypothesis Watch

(15:58) School Spending

(18:06) Costs Are Not Benefits

(19:08) Fire The Bad Teachers And Hire And Retain Better Ones

(19:18) Teachers Unions Are Not Beating The Rumors

(21:11) Death Spiral

(22:07) Another One Bites The Dust

(24:13) Where To?

(24:59) Public Schools Cost A Lot Of Money

(27:04) Schools Can Fail In Many Ways

(28:55) Results

(30:25) Grades

(30:42) They Nerfed AP Tests

(31:28) What Happens Without Academic Standards

(37:10) The Two Big Scandals

(38:58) Alpha School

---

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.

Anthropic filed its S-1 this week to prepare for its IPO. They still have to fill a lot of the numbers including the share price. That still left a lot to be learned, including how the messaging to investors differs from their other rhetoric, and exactly how their corporate control is structured. Anthropic remains a frustrating company, doing a variety of things that often seem in tension with each other.

Google is claiming to be fully back, with the announcement of a new model Gemini 4 Carbon that tops the new AA v5 benchmark index. It looks to be a large leap over Argon only two weeks later. No, you cannot use it. Google fails marketing forever, so they give us the announcement and then weeks later you might be able to enter some queries. That largely invalidates my look at the Argon model and safety cards, but it can at least form a basis for looking at Carbon as well.

Many eyes remain on The Hackening, as additional things get hacked. So far most of the important hacks have been by read teamers and white hats, so there has not been major damage. We are playing [...]

---

Outline:

(01:53) Language Models Offer Mundane Utility

(04:06) Language Models Don't Offer Mundane Utility

(04:40) New Math

(07:25) Huh, Upgrades

(09:06) Artificial Analysis v5

(17:27) On Your Marks

(18:34) Gemini 4 Argon Was a Good Model But Not Fully Frontier

(20:27) Gemini Carbon: Is Google Back?

(21:56) Get My Agent On The Line

(24:15) Spy Versus Spy

(25:40) The Hackening

(35:13) Slack Off

(43:20) Help Peer

(49:14) Fun With Media Generation

(50:09) Copyright Confrontation

(50:28) You're Fired

(55:50) Verify the Software

(57:06) They Took Our Jobs

(57:13) Get Involved

(59:19) Introducing

(59:40) In Other AI News

(01:00:20) Show Me the Money

(01:00:46) Bubble, Bubble, Toil and Trouble

(01:01:27) Quiet Speculations

(01:06:58) The Quest for Sane Regulations

(01:07:23) Chip City

(01:10:40) The Week in Audio

(01:10:59) People Just Say Things

(01:14:22) Rhetorical Innovation

(01:18:43) [Artificial Intelligence]

(01:19:34) Misaligned! (At Alibaba)

(01:28:04) Stop It

(01:29:57) METR Red Teams OpenAI Internal Agent Monitoring

(01:33:13) Massive Fraudulent Distillation

(01:34:04) Aligning a Smarter Than Human Intelligence is Difficult

(01:37:56) OpenAI Attempts to Secure Its Frontier Model Training Environment

(01:46:36) The American People Really Hate AI

(01:47:03) The Lighter Side

---

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.

The full text is finally here, so we are giving it the full RTFB of model card treatment.

First, a recap of the state of Anthropic finances, then a close reading of the relevant parts of the S-1. A lot of it is technical and uninteresting, often pro forma, and gets skipped. A lot of it gets repeated many times.

A lot of this is also Anthropic ‘playing the hits’ we have heard many times, now to the investor circuit. Get your drinks ready for when we have to maximize the upside and minimize the downside.

This presentation of S-1 data has some cool Claude-created charts if that's your jam.

Here are my top 10 takeaways from the S-1, for those already familiar with Anthropic.

---

Outline:

(02:48) Show Me the Money

(10:11) Anthropic Has The Money

(11:47) Shares of Many Classes

(13:57) Who Governs Anthropic?

(18:57) The Cap Table

(19:43) Realize the Benefits While Navigating and Preventing the Risks

(20:42) Just Think of the Potential

(21:20) People Are Worried About AI Killing Everyone (1)

(23:25) Anthropic Bets Big on AI

(24:42) Anthropic Does Not Fully Maximize Profits

(27:48) Everything Is Awesome

(29:23) How Much Is AI Worth?

(30:37) How Much Would You Pay?

(31:48) Minimize Those Risks

(33:58) Hopefully Introducing the World's Most Powerful Model

(38:25) Might Not Compute

(40:11) Supply Chain Risk (1)

(40:32) Anthropic Is Nothing Without Its People

(45:00) Shitloads of Money

(46:40) Someone Has To Buy The Product

(47:51) Did You Know There Are Other Frontier Labs?

(48:27) Why They Hate Us

(49:28) Claude Seems Pretty Misaligned to Anthropic

(53:42) People Are Worried About AI Killing Everyone (2)

(59:26) Someone Might Sue or Regulate Anthropic

(01:00:38) Supply Chain Risk (2)

(01:01:07) We're Too Safety

(01:02:19) Claude Core

(01:02:41) We've Never Actually Made Money

(01:04:06) You Are Buying Stock In a Company

(01:05:04) There Will Be No Dividends

(01:05:38) What Are We Even Doing?

(01:07:18) The Financial Model

(01:11:55) The Business of Intelligence and Maybe Killing Everyone

(01:15:10) Don't Give a Damn About My Regulation

(01:17:08) The Bottom Line

---

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.

Anthropic is a frustrating company.

Sometimes it will do things I greatly appreciate, including at substantial cost. They have taken bold stands, have a unique governance structure and have hired and worked to establish a relatively strong (but I think still inadequate) safety culture.

Other times, it will do things that are deeply disappointing.

It produces very good models. That too is not entirely a good thing.

One can both be the best given the alternatives, and insufficient to meet the moment.

Ultimately, what you think about Anthropic depends on what parts you care about most, and what you think would be sufficient.

Anthropic is great if you think the problems are important and difficult, but within the range of what Anthropic's approaches can solve. If you think we mainly need operational excellence and caring about safety and the future, then there are some non-excellent operations to fix but that goal seems doable for Anthropic. If you think the problems go well beyond that, then things look rather grim for everyone.

Today's post begins with a continuation of discussion over Anthropic's leaked draft bill. See my previous coverage for details of the [...]

---

Outline:

(01:43) Dario Amodei Responds to Nathan Calvin

(20:50) Kelsey Piper Goes Inside Anthropic

(26:19) The Technical Problems of AI Safety Are Not Solved

(27:50) The Quest for Embedded Evaluators

(33:57) METR Questions Risk Report Methodology But Affirms Risk is Low

(38:58) The Anthropic Public Benefit Fund

(41:39) Anthropic Versus Anthropic

---

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.

Gemini 4's model card is a compact 14 pages. Its safety and security report has 217.

This is The Way. It also means we have a lot of new reading material.

I will quickly summarize the model card, then focus on the safety card.

As always, many statements here also apply to other frontier labs.

A lot of this one is probably for diehards (and members of technical staff) only, plus those who find this interesting. Most of you do not need to dig into the cards on this level, and I found most of what we see entirely unsurprising.

The biggest takeaways are:

---

Outline:

(03:02) The Model Card

(04:31) Safety Card Executive Summary (1)

(07:30) Safety and Security Mitigations (2)

(10:11) Frontier Safety (3)

(11:52) The CCL and TCL Thresholds

(14:56) Biological Misuse (3.2.1)

(19:38) Refusal Rates

(20:43) Chemical, Nuclear and Radiological Misuse

(22:01) Cybersecurity (3.2.2)

(24:45) Harmful Manipulation (3.2.3)

(25:40) Misuse Safeguards (3.3)

(26:30) Loss of Control (3.4)

(30:23) Either AI R&D Is Accelerating Or It Isn't

(32:22) The Holistic Loss of Control Assessment (3.4.3)

(35:49) Alignment Assessment (3.4.4)

(42:41) Chain of Thought Transparency (3.4.5)

(47:32) Voices in My Head

(48:48) I Can See Everything

(53:06) Security Mitigations (3.6)

(54:14) Preventing Content Safety & Agentic Misuse (4)

(56:50) Model Distillation (4.3)

(57:36) Promoting Safe & Aligned Model Interactions (5)

(59:48) Sycophancy (5.4)

(01:01:28) Hallucinations (5.5)

(01:01:51) Helpfulness (5.6)

(01:03:11) Agentic Robotics, Wait What (5.7)

(01:04:31) Preventing Harm to Users from Third Parties (6)

(01:05:31) Strengthening Cyber Defense Capabilities (7)

---

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.

And So It Begins: PHP

---

Outline:

(01:12) And So It Begins: PHP

(05:19) The PHP Bug Was Kind of a Big Deal

(09:45) On Your Marks

(12:25) This Was Only the Beginning

(16:01) Some More Hacks

(17:56) Everything Will Be Attacked

(22:22) Hugging the Face

(24:49) Get Daybreak Now

(26:15) Some People Need Practical Advice

(30:51) Diffusion Is Slow Until It Isn't

(33:11) One, Two, Three, Four

(34:36) Going Vertical

(35:22) Bot Versus Bot

(36:28) Cyber Lack of Security

(39:09) Offense Versus Defense

(42:11) Pick Up the Phone

(45:57) What Can We Do About It?

---

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.