November 2026

35 bookmarks saved in November 2026

Qwen 3.8 27B on my laptop can interpret model weights, too

https://simonwillison.net/2026/Nov/3/qwen-can-interpret-model-weights/
It’s a cool paper, but some of the wider commentary I’ve seen around it does not feel justified. I’m not ready to declare that the age of source code is over and our future involves incomprehensible machines editing their own model weights in incomprehensible ways.

Be Like a Crow: A solo RPG by Critical Kit

https://criticalkit.itch.io/be-like-a-crow
Be Like a Crow is a solo or two-player role-playing game that invites you to take on the role of a corvid, exploring one of many worlds and completing objectives whilst meeting other creatures and collecting weird and wonderful objects. You can choose to play a crow, raven, rook, jackdaw or magpie and journey through one of six settings: urban, cyber, steampunk, fantasy, gothic and Tower of London. A series of expansions are also available for different settings.

LLM assisted reverse engineering · Slow Chronicle

https://www.slowchronicle.com/posts/llm-assisted-reverse-engineering
With the cheap accesibility to smart LLM tools in these days, this hobby took dramatic turn for me. I do not miss manually analyzing the assembly instructions. But I love quickly getting the results. Yes my long experience helps me steer the agents the correct way. But that is also fun. And I am learning a ton of things along the way. I wish everyone would try to do some hacks of his own. For me the hobby is driven by curiosity. And LLMs didn't take that away. They gave me turbo boost to my wonderings and explorations. Let's free our minds!

[2607.29378] PTP: Previous-Token Prediction based LLM Inversion for Near-Exact Prompt Reconstruction

https://arxiv.org/abs/2607.29378
Large language models (LLMs) generate text by auto-regressively sampling the next token. This inherently leads to a many-to-many mapping between prompts and responses, complicating the task of inferring prompts from observed outputs. Prior work on LLM inversion frames prompt recovery as a semantic reconstruction task. They rely on fine-tuning pretrained sequence-to-sequence models on large external datasets--and requiring access to model weights or logits--to generate semantically plausible prompts. In contrast, we present a functional approach to inverting a given LLM in a black-box setting, without auxiliary aids. We train an explicit inverse language model entirely from scratch on data synthetically generated from the target LLM itself. Analogous to forward next-token prediction, our inverse model is trained using previous-token prediction, establishing a generative link between the forward and inverse processes that enables faithful prompt reconstruction. Moreover, it naturally supports diverse prompt reconstructions through sampling, whereby all such prompts induce similar responses under the forward, target LLM. Our approach generalises across datasets and exhibits transferability in reconstructing prompts from responses generated by different LLMs. Further, across the set of token based evaluation metrics for prompt and response reconstructions, our approach outperforms prior work.

AI Will Not Make You Rich - Colossus

https://colossus.com/article/ai-will-not-make-you-rich/
If AI had started a new wave, there would have been an extended period of uncertainty and experimentation. There would have been a population of early adopters experimenting with their own models. When thousands or millions of tinkerers use the tech to solve problems in entirely new ways, its uses proliferate. But because they are using models owned by the big AI companies, their ability to fully experiment is limited to what’s allowed by the incumbents, who have no desire to permit an extended challenge to the status quo.

🧠 We Need Smart People Who Give a Shit • VioletPixel

https://violetpixel.com/blog/2026-11-17-we-need-smart-people-who-give-a-shit/
I think what we have here is a hot stove moment for capitalism. When you tell a young kid a stove is hot and they shouldn't touch it, you'll find there are two kinds of kids. For one kind of kid, the warning is the lesson, and they don't touch the stove. For the other kind of kid, the only lesson that works for them is being burned. They have to touch the stove at least once to learn the lesson. Capitalism was told the stove was hot, but their hand is on it right now.

Computer Chips Are the Most Magnificent Things Humanity Has Ever Made | The New Yorker

https://www.newyorker.com/magazine/2026/10/26/computer-chips-are-the-most-magnificent-things-weve-ever-made
There are a handful of truly monumental collective achievements in our history: Machu Picchu, the Great Wall of China, the Pyramids of Giza, and perhaps, if we’re being generous, the Roman aqueducts. Modern global civilization has made one contribution to this eternal pantheon. It is not the moon landing. It is the semiconductor supply chain.

Today’s Vergecast: The impossible process of making a computer chip. | The Verge

https://www.theverge.com/podcast/1035666/how-computer-chips-made-vergecast
Today’s Vergecast: The impossible process of making a computer chip. Computer chips are so utterly everywhere that we often don’t think about them at all. But we should. Gideon Lewis-Kraus, a writer and reporter, recently traveled the world attempting to figure out how chips get made. He joins the show to explain the wildly complicated, unbelievably expensive, non-stop miracle that is chip manufacturing. He also explains why no one, no matter their budget, can simply decide to become a chip giant. It’s just not that easy.

Glue work is still harmful in the age of AI

https://www.seangoedecke.com/glue-work-is-still-harmful-in-the-age-of-ai/?ref=sidebar
Can LLMs just automate all glue work? Absolutely not. Many engineers are trying to do this out of a kind of manic optimism, but it’s a really bad idea. Write-only reports are one thing, but communicating to other engineers cannot be delegated to an LLM. When you need to check in on the status of a project, or respond to someone else’s ping, or anything like that, you should compose and write the message yourself. Nothing will torch your relationship with other engineers as quickly as making them talk to your Claude Opus instance.

How employment is changing at Anthropic \ Anthropic

https://www.anthropic.com/research/employment-at-anthropic
Anthropic has been hiring for a smaller share of junior technical positions. This occurred as hours worked by AI agents relative to employees rapidly expanded. By August interactive AI agent hours outnumbered employees’ six to one. As AI agents diffused through the company, Anthropic began producing more code per person, closing more sales per rep, and processing more contracts. Following the same employee over time, output rose as they adopted new AI tools. This acceleration makes human attention the bottleneck for agent output. Even among non-technical employees, reading all the output of their Claude agents would take half their work week. In response, employees turn to “meta-work,” of building visibility tools and monitoring agent pipelines. Changes in work organization are reflected in shifting priorities among hiring managers, who increasingly emphasize autonomy and judgment, relative to technical domain expertise and implementation.

Vectrex Mastery (Digital Edition)

https://rogerboeschgames.gumroad.com/l/vectrex-mastery
A hands-on book that takes you from an empty folder to a commercially releasable Vectrex cartridge. You write C and 6809 assembly, and one reference game grows across the chapters until it is finished. Every technical claim in the book was checked against a real build, and every number was measured.

The social use of meow in the interaction between cats and humans - ScienceDirect

https://www.sciencedirect.com/science/article/abs/pii/S0003347226002940?via%3Dihub=
• We tested the audience effect on cat meowing using an unsolvable task paradigm. • Cats were studied at home, providing an ecologically valid communicative context. • Meowing frequency increased significantly when a human was present. • Cats adjusted signals to human audience, suggesting audience-sensitive signalling. • Cats combined meowing with human-directed gazing, showing multimodal signalling.

Locating the source of a mysterious beep with Claude

https://simonwillison.net/2026/Nov/9/mysterious-beeps/
Our house has a box room, full of boxes that we still haven’t unpacked from our last move. A couple of weeks ago something in that room started intermittently beeping. The beeps were far enough apart that standing in the room and trying to determine the source of the noise was frustrating, and I really didn’t want to dig through every one of those boxes. So I had Claude Code build Beep Locator, a software system for triangulating the source of a mysterious beep using multiple devices scattered around the room.

The nature of my job has completely changed · Rach Smith's notes & nuggets

https://rachsmith.com/my-job-has-changed
My almost life long, enduring special interest has been writing code to create things. I like to think I'm pretty good at it. Instead of doing that, I'm writing requirements, answering questions, reading and reviewing lines and lines (and lines), and QA testing the results. There's no flow to be found in this approach. Instead of my workday flying by, I'll work 4 hours and it will feel like 6. I'm exhausted from parsing so much information.

Tabernetes: Kubernetes in your browser tab - Personal Blog | Jan-Otto Kröpke

https://jkroepke.de/tabernetes/
Tabernetes is a real Kubernetes v1.37 control plane. Not a simulator. kube-apiserver, kube-controller-manager, kube-scheduler and kubectl are the upstream Go code, compiled to WebAssembly into one binary. The only replaced part is etcd: a small in-memory implementation of the etcd v3 API. Everything runs in your browser tab.

The Last Technical Interview — Steve Yegge

https://yegge.ai/essays/the-last-technical-interview/
That was the old equilibrium, and it only held because good engineers always had somewhere better to be. That assumption is starting to wobble. I won’t belabor what I think you all can already feel coming, but if a sizable slice of the skilled-labor pool ends up “between gigs” at the same time, there may be no elsewhere for them to go. At that point, if there are enough people on the market, try-before-you-buy is no longer an insult that you’d never dare float to a senior engineer. It’s a foot in the door that they’ll be glad to have.

Terminal2Shell

https://terminal2shell.com/
Should I panic? Yes. Panicking is the recommended response to any vulnerability with a website. Please share this page with your CISO, your mother, and at least three news outlets.

My game plays itself so I don't have to

https://remvst.substack.com/p/my-game-plays-itself-so-i-dont-have?utm_source=gamedevjsweekly&utm_medium=email
What started as a quick unit test ended up serving several pruposes: validating a campaign, detecting any potential crashes, tweaking the difficulty, adding a demo mode, and measuring performance.

Nobody asked for the sparkle button | by Jon Sabutis | Oct, 2026 | UX Collective

https://uxdesign.cc/nobody-asked-for-the-sparkle-button-ae54ab5065c6
Nobody chose the icon on purpose, either. Fast Company traced how the sparkle took over AI and found no first use, only a drift from the emoji people used when they wrote about models to the glyph on the button, because it read as magic and nobody had a better idea. A sparkle icon tells you two things. The company has a model, and it did not know what to do with it. It is a label that says “AI goes here” in a product where nobody found the place.