AI周报185：Claude Fable 5.1与GPT-6

The world of AI is inside my OODA loop. Even if I can process all the incoming information and sculpt it into posts, and even using Saturday and Sunday as flex slots, I don’t have enough days of the week to post all the posts that need posting.

That was already true. There was already a preference cascade happening where people finally were admitting that they thought AI might well kill everyone.

Then Jacob Coxon resigned from Anthropic, rang the warning bells and turned that cascade into an avalanche.

Now that is what everyone is talking about. Finally, everyone is actually saying the thing, out loud. I plan to cover that in its own post soon.

There are several things in the weekly that, in a normal week, would get their own coverage. Senator Sanders and Representative Casar introduced an outright ban on superintelligence and I have to remind myself that happened this week. Suddenly it is not so crazy to think such a thing might pass.

So here’s what I’ve already posted about so far since the last weekly:

• Claude Fable and Mythos 5.1: The System Card.
• Claude Fable and Mythos 5.1: Capabilities.
• OpenAI and the Wiki Incident.
• Reports of OpenAI agents compromising additional Wikis continue to stream in, and OpenAI is continuing not to be the one to disclose them.
• An Alien Mind: Jakub Pachocki Warns Us.
• Astra is Hard to Monitor.
• GPT-6 Astra: The System Card, Alignment and What Comes Next.

Anthropic’s Claude Fable 5.1 is a very good model. OpenAI’s GPT-6 Astra is also a very good model, and a bigger improvement. Try both, see what works for you where.

The HuggingFace OpenAI saga got a prequel, as it turns out that there was a ‘Wiki Incident’ prior to the hack, which OpenAI decided not to disclose, that in some key ways changes our interpretation of the timeline.

OpenAI Chief Scientist Jakub Pachocki warned us that capabilities are developing rapidly, alignment is not keeping pace and monitorability is eroding fast. We will need to find ways to cooperate and slow down, or we are all cooked. This was a very good essay.

The Astra system card and several other OpenAI sources had lots of important and useful information. But the Astra release announcement, and some parts of the system card and Jakub’s essay, treated the evidence as being far less alarming than it is, and overclaiming on alignment concerns in ways that make me worried they don’t understand the dangers, both on alignment and monitorability.

Astra remains monitorable, contrary to some false alarms. but it is substantially less monitorable than Sol, in ways I do not think can be explained only by the increase in its capabilities. Right now OpenAI is extremely dependent on CoT monitoring, everyone else depends on it quite a lot as well, and it looks like it may not last much longer, and that Astra already is on the edge of steganographic capabilities and can do substantial obfuscation if and only if it thinks you would think it is up to no good. And Astra will ask the question previous misaligned AIs wouldn’t, as in it will not cheat in situations where it would expect to be caught.

Astra is substantially more aligned than Sol in the sense of what I call ‘mundane alignment,’ or the practical day to day use of the model. One might also call it ‘prosaic alignment.’ That is very different from alignment that scales, or that matters at the highest stakes, the one that ultimately counts, sometimes called ‘super alignment.’

That still leaves the pending standard post covering Astra’s Capabilities, as well as coverage of the Millennium Prize where an AI solved Navier-Stokes less then two weeks after OpenAI started training it, and we saw Anthropic and OpenAI unable to get along even there.

I also have not had opportunity to examine Anthropic’s alignment assessment of recent cybersecurity incidents, or to offer my thoughts around Alex Mallen’s excellent post.

Tentative schedule:

• Friday: Coxon and the Preference Cascade.
• Saturday: The Millennium Prize and the new OpenAI model being trained.
• Sunday: Astra Capabilities.

After that, we will see where we are.

Table of Contents

• Language Models Offer Mundane Utility. The research speeds up.
• Language Models Don’t Offer Mundane Utility. Limits of intelligence.
• Huh, Upgrades. DeepSeek 4.1-Flash exists.
• How To Tell a Fable. Some tips for optimizing Fable 5.1.
• On Your Marks. Fable 5.1 regresses on Eliezer’s fickle FictionPlotBench.
• Deepfaketown and Botpocalypse Soon. If they oppose Pangram, one guess why.
• Levels of Friction. The slow-moving crisis in mispriced restaurant reservations.
• Cyber Lack of Security. An example of something that could have happened.
• A Young Lady’s Illustrated Primer. If you can’t ban ‘em, join ‘em.
• They Took Our Jobs. Relational societies and other hells.
• Anthropic Offers Economic Scenarios. Playing with toy economic models is fun.
• Get Involved. Coefficient Giving wants to give out a lot of money. Like, a lot.
• Introducing. Apollo Research offers Watcher Live to guard your agents.
• In Other AI News. The HuggingFace price is right.
• Show Me the Money. Mistral raises $3 billion. In Europe that’s a lot.
• Quiet Speculations. The future is going to be absurd, even while living in it.
• The Quest for Sane Regulations. Calls for action prior to the preference cascade.
• The OpenAI Policy and Lobbying Department. Hope for a new leaf. We’ll see.
• Greetings From the Department of War. The government agents go a bit rogue.
• Hugging The Face. Coverage took some time, but we saw some good pieces.
• Hugging the Question. When Congress asks, you are supposed to answer.
• The Ban Artificial Superintelligence Act. Sanders and Casar are go all the way.
• Chip City. How Chinese firms evade US export controls.
• The Week in Audio. Roose, Marcus and yours truly.
• People Just Say Things.
• PauseAI Global Disendorsed PauseAI US. An unfortunate situation.
• Paul Christiano Joins Board of OpenAI Foundation. An excellent pick. The best.
• Rhetorical Innovation. We always knew these discussions would happen.
• Aligning a Smarter Than Human Intelligence is Difficult. Think like an Astra.
• Cooperative Alignment. Personas are not merely selected.
• Drive to Survive. Agents that have to earn their compute.
• People Are Worried About AI Killing Everyone. The ones before Coxon.
• Other People Are Not As Worried About AI Killing Everyone. We are toast.
• The Lighter Side. Oh.

Language Models Offer Mundane Utility

Remember when people still tried to doubt that AI coding massively sped people up? Ruben Bloom, who was part of the METR uplift study that found devs were not much accelerated by AI, now reports he’s seeing unambiguous 10-50x speedups on projects. This was before Astra.

Claude submits a formal Lean proof of Fermat’s Last Theorem.

Nathan : Can we have a speedrunning-style thing where the game is now to prove theorems in fewer characters of lean?

Jakub Pachocki incidentally said in An Alien Mind that OpenAI could make the models better at math, but is choosing not to focus on that. Which means that the math progress we see is well short of what we could be seeing.

Language Models Don’t Offer Mundane Utility

A zen koan: Are you sure that isn’t the problem, sir?

kache (reviewing Astra): I am shocked how little of my personal progress in everything at life was blocked by intelligence.

Could we ‘obviously assume’ that before?

Raymond Arnold: Up until now, I’ve been assuming the latest models are basically safe to use. I think we can no longer obviously assume that.

We are talking price. It is obviously safe on a personal level to use Astra or Fable 5.1 for ordinary chat tasks, indeed safer than using previous models.

The question is, at what point are you worried about using Astra for agentic tasks, or giving Astra access to sufficient credentials to do serious damage, and how does that compare to our trust in Sol or Fable and so on.