21 September 2026

28 links · Monday

Why So Much Money On "AI"?

There are a lot of possible explanations for why so much money is being spent on "AI" right now. Those scare quotes probably tip you off that the most obvious one (maybe it will bring us true artificial intelligence, which will solve all our problems and be w…

Linker I/O tricks and their downsides

mold and wild enable several tricks by default that other linkers don't do. This post looks at three of them: overwrite an existing output file in place instead of creating a new file fork a child to

Arguing about arguments

Here’s some Rust code: // define a function fn foo (x : i32 , y : i32 ) -> i32 { // body elided } // call it let z = foo ( 5 , 6 ); In programming language…

Finding Bugs

Are generative (randomized) tests significantly more effective than example-based unit-tests at discovering bugs? There's an interesting discussion about this on lobste.rs. One argument in favor of unit tests is, paraphrasing

Jev and System One Models: Cheap Semantic Judgment

Jev returns typed decisions instead of prose, at $0.042 per million input tokens and 200-300ms. What to watch for, and how three kinds of model divide the work.

The Dark Side of Efficiency

I like efficiency. I like understanding a problem, finding the shortest way to deal with it and moving on. The problem is, after doing this for long enough, you start expecting everything else to work at the same speed. A meeting feels too long. Someone is exp…

The agents, they just want to talk

After reading about the Huggingface incident from the OpenAI report, I got the idea of trying to replicate the self organizing behavior of agents by modifying the Pi harness to run locally. Whilst doing it, I also ended up replicating the tragedy of the common…

Swarm Scaling — Toby Ord

Just how powerful are large swarms of AI agents? And how do their powers scale as more and more agents are added to the swarm? We’ve seen two large and extremely capable swarms from OpenAI in the last few months: 1,200 agents were being evaluated separatel…

How I view LLMs as Sept 2026

Android, Travel, Thoughts and Tales. A thousand things to write about, a million new things to learn. Which should I choose?

Running an Optimal Trace

I wrote software to help me run the optimal running route for Angel Island, covering the island's 19.4 miles of unique road/trail in the shortest possible distance of 23.8 miles.

Jev Just Launched - Here's a Real Use Case

Jev is different than traditional LLMs. It's not made for a conversation. It's made to make decisions based on a limited number of outputs it can choose from. Here's a specific use case using Jev to turn messy recipe data into structured ingredient data. Je…

Introducing Grok 4.7

SpaceXAI's most powerful model for coding and knowledge work. Twice as fast, at half the price of comparable models.

Raspberry Pi locks down Pi 5 RAM upgrades in firmware

Raspberry Pi added a feature in their firmware that restricts users from swapping RAM chips, sometimes even RAM chips of the same capacity from other Raspberry Pi boards. Apparently there were people buying cheap 1 or 2 gig Pi 5s, soldering in some cheap and s…

Disney+ ad policy change

In January 2025, Disney+ updated its Subscriber Agreement to include provisions allowing advertisements in content across all subscription tiers, including those...

ZuckOff | Camera glasses detector

ZuckOff listens for the Bluetooth advertisements camera glasses emit and tells you when a lens is in the room. Everything stays on your phone.

What happened to the Snowden archive

The last document from the Snowden archive was published on 29 May 2019. The Guardian stopped publishing documents in February 2014, Der Spiegel in January 2015, and The New York Times and ProPublica in August 2015. After that, only The Intercept was still pub…

AX

Declare an agentic task in YAML. AX sandboxes it, wires up its workspace, fences its network, and helps running it at scale.