I have asked both fable 5.2 and opus 5.5 to recreate docker logo redesign promotional video.
Which model won?
A new Docker, but a familiar story.
Today we're introducing a new look, a new focus, and a refreshed Moby.
Same Docker you've trusted for over a decade, now built to run on trust in the agentic era. We can't wait to show you.
Read the full story: bit.ly/4iX61ek
1
25
Borouch retweeted
Today’s claim on the Riemann Hypothesis helps demonstrate what I mean when I speak about Computational Reenchantment.
If the proof holds, we’ll likely simplify its explanation to some extent.
But as it stands, it’s 250,000+ pages.
Long enough to indicate that “pages” is a vestige of a mode of formal comprehension we’re rapidly leaving behind.
So really no one, including the proof’s human “author”, comprehends the nature of the proof at full resolution.
But because computational verification scales far more steeply than human comprehension, and because it can offer certain functional guarantees, we will use abstractions we comprehend less and less to do more and more.
At some point this process will produce physical creations indistinguishable from magic and the paranormal, even to their creators.
As we push into the deepest realms of analytic formalism, the phenomenological bottom drops out, leaving us once again grasping for understanding amidst an age of generally inexplicable Angels and Demons.
We are multi-agent with different goals and valus with varying degrees of alignment, buddishm no-self
.@jkcarlsmith (Anthropic) explains why you lose part of your brain when you have major internal disagreements
"Doing work that you don't actually want to do, there's some part of you that is like, "I'm not bought into this project. I'm out."… you're losing a part of your brain… being real, authentic, sincere, something is getting all parts of your cognitive resources to be together and look at each other and be willing to be in the same room and doing something together. And if half of them are noping out, or half of them are pretending they're not there, and half of them are going to sleep because they're bored… then you're less alive."
8
Ideas are animate objects distributed across time and place
They embed and prolifirate inside the minds shaping the societies and individuals
They should be viewed as meta objects in itself to be studied
Abstractions have causal power. This is the key to understanding everything from religion to biology to computer science. It is a great and terrifying idea.
6
Models need high entropy sources which take them out of loops and average distribution, probably half my skills are about that
Replying to @fchollet
I think we need to inject creativity and entropy into the process somehow.
I ran it to optimize a quest script in 2004 RS and most of its own ideas were incremental, all the most dramatic speedups came when I directed its attention.
xitter.gq/RobertHaisfield/status…
1
30
Soon a company will emerge which will hire people solely on reading threads of person interacting with agents.
5
If you look at meta level. The fact of llms arrival should itself produce breakthroughts in philosophy of mind, epistemology, linguistics, interpretability, computation, animism and so on, just not enough people integrating these ideas
1
44
Or hooks. E.g a hook to reject commit if it hasn't been reviewed by subagent
The best way to keep clanker slop at bay is aggressive linting.
You can write until you're blue in the face about how you want stuff to be done in prompts and instructions, but they always go their own way.
But they always listen to linters. Put everything into linting.
8
Markdown harness
Code skeletons (types, modules, signatures) beat prose design docs and behavioral specs in AWS’s new Skeleton Coding paper.
- Code skeletons raised SDDBench mean test pass rate from 29% to 74% vs a task description alone.
- With skeletons plus global invariants, a smaller model came within 5–8 percentage points of the larger model’s best results at 15–26× lower token cost.
The best-performing specs combined modules, types, and signatures with natural-language contracts and global invariants. Richer .md architecture docs helped, but remained below the best code-like skeletons. Excessive contract detail hurt performance, especially for smaller models.
Skeletons also reduced code erosion (a complexity-based proxy for maintainability) while behavioral specs increased it.
Theory: structure changes which tests are sufficient. Interfaces expose intermediate states that end-to-end tests can miss.
Limitations: the skeletons came from reference implementations. Costs exclude specification generation and evaluation.
Unclear: how cheaply we can produce this structure upfront, and whether making the contracts mechanically checkable would improve results enough to justify the added cost.
My bet: the best starting point for a coding agent isn’t more prose. It’s a model of the system and explicit constraints the implementation must satisfy.
arxiv.org/abs/2610.38214
18
Markdown harness > wiki
Knowledge must flow.
By modeling information pathways, setting up reflection loops correctly models can have excellent memory and an increase in performance asif they from next generation
i have been using a wiki system for a few months. benefits:
1. wikis have preexisting format/best practices for complex topics, so you largely don't have to define rules yourself
2. wikis naturally grow outwards. agents can autonomously detect gaps in information and fill them in
15

















