Notes on Dan Shearer

Structure vs Constitution in AI Safety

Tue, 31 Mar 2026 20:00:00 +0000

Anthropic publishes its constitution along with research about where the constitution works and where it does not. The current version is an ethical treatise addressing Claude discussing safety, ethics, Anthropic’s guidelines, and helpfulness, in that order when they conflict. Anthropic favours cultivating good values and judgment over strict rules.

In 2026, Anthropic’s operational judgment failed twice in the same way, leading to the leak of the Claude Code source code.The constitution asks Claude to imagine how a “thoughtful senior Anthropic employee would react”, but what happens when the organisation’s structure fails?

Snow Crash and Standing Orders

Mon, 30 Mar 2026 19:00:00 +0000

I am using my personal Perseverence engine as I help develop the code, and I’m watching carefully to see how useful it is for developing analysis, review and writing. Evidence so far is mixed, but improving fast. I feel in control as I do with any other work tool, which I certainly do not when using a typical error-prone AI text interface. One of the reasons I feel in control is because there are more controls in place, that is the point of the Artificial Organisations concept. But another reason is that this tool is becoming more tuned to me all the time.

AI, PCE and the Geth Consensus

Fri, 06 Mar 2026 06:55:19 +0000

AI ethics and safety work mostly focuses on making individual models smarter and better-behaved via guidelines and persuasion, with not much hope this will succeed. You should especially not feel safe when an AI company reassures you about their guardrails. When you hear guardrails think of telling a dog “Don’t bite the furniture inside the house today”, because you can never know what will actually happen. The concept of Artificial Organisations doesn’t require AIs to be reliable, it ensures that when an AI goes wrong there are hard limits on how much damage it causes. Similarly we can put the dog outside the house, so no matter how bitey it is the furniture cannot be bitten. I have been spending a good deal of 2026 trying to use this concept to make AI less dangerous and more useful. I even have it studying me as an apprentice. This is mostly the opposite to Anthropic’s idea of a constitution.

Patents and the MIT License

Tue, 10 Feb 2026 14:30:00 +0000

Patents and the MIT License

Some of my software projects use MIT so I have studied this issue. Although in many respects the world has moved on from copyright wars to much higher-stakes legal shenanigans, the detail of licensing still matters.

In my case:

My LumoSQL project is based on probably the most-used software, SQLite, whose license states it is in the “Public Domain”. The meaning of this isn’t entirely clear in some cases, and a 21st century software project starting decades after SQLite shouldn’t copy this.
I chose MIT as a commonly accepted alternative, but which license is that exactly, and what does the text imply about patents? This is known, but I had to dig.
The MIT license is massively used, but who will defend it if needed? We know the answer for the GPL, and also Apache-type licenses. I am now satisfied that quite a lot of enormous organisations really do care about MIT.
There are lots of reasons why MIT isn’t ideal, but in my view those are trumped by it being widely accepted as fit for purpose, and relied upon by organisations who care that it remains effective and unambiguous.

My notes are mostly kept in my many contributions to the Wikipedia page on the MIT License since that is where the decades-old knowledge of the MIT license origins is already maintained. The legal minds in many of the largest companies in the world seem to accept that at least in the US the MIT license implies a patent grant. As probably the most-used open source license, the MIT license has many wealthy corporate defenders if anyone wanted to test that idea.

How this site is made

Tue, 10 Feb 2026 12:00:00 +0000

I made a new website recently. My goals:

Modern-looking
Easy to maintain, minimal infrastructure
Content lasts indefinitely even as web technologies come and go

I decided on a static website, with content in Markdown and a modest amount of templating. I chose the Hugo static site generator with the PaperMod theme, plus a second theme for CV-type timelines. I used bundled system fonts (no Google Fonts tracking by calling googleapis).

I added small customisations using CSS and Hugo shortcodes including colour themes, a general timeline (in addition to the CV one), handy infoboxes and the like. Hugo makes this quite easy to achieve while still using mostly standard markdown. That bodes well for being able to move to other systems as the years roll on.

Website challenge

Tue, 10 Feb 2026 12:00:00 +0000

My new website is nice enough, but it really needs work. I’m offering prizes!

Small fixes for wording, grammar or links — my warmest thanks
A page or more of such small fixes — I will buy you the (non-outrageous) beverage of your choice
A substantial improvement or correction consisting of a page or more — a pizza from a mutally agreed place
10 non-trivial pull requests for the codeberg repository — I’ll help you learn Linux, if that’s a thing you want
Assistance to help me fix items from the following list — prizes as per the above, based on scale/complexity

Things to be added or improved:

A Design Challenge for Horologists

Wed, 04 Feb 2026 09:00:01 +0000

I have an interest in non-electronic computers as an educational tool. An horologer is someone who makes mechanical clocks and watches, and horologers definitely don’t believe in electronics. That’s why I published Shearer, D. (2007). "Communication: A Request for Collaboration." Horological Journal, 149(12), p. 471., which sounds much fancier than the letter to the editor it was. Given I didn’t even know the British Horological Institute existed until a week prior it makes me very pleased.

Open Source to Chemical Rockets

Mon, 02 Feb 2026 05:55:38 +0000

(written in 2008)

How a young Australian discovered Open Source and a career. Eventually learning that a mixture of code, law and mathematics is a frontier for human rights battles.

It isn’t often I come face to face with myself after a twenty-something year break, but I did yesterday.

As a first year university student at the South Australian Institute of Technology in Adelaide I did landscape gardening oddjobs for companies. I noticed a company called Australian Launch Vehicles (ALV), which sounded very cool, so in I went. ALV was founded by a pair of entrepreneurial rocket scientists. Despite decades of rocketry history in South Australia, there was no local space industry. (Establishing Australian spaceflight in 1987 was ambitious; they failed but others are giving it a go.)

Radio Waves to Random Number Generator

Sun, 11 Jan 2026 00:00:00 +0000

Random humans and computers

Humans are terrible at randomness. If you ask people to write down a list of random numbers the result can usually be shown to not be random at all. Stage magicians and marketing experts exploit our inability to assess how random an event is.

But computers, surely they should be random? It sure feels like it when your printer jams. But no, computers are often worse than humans at being random, and that’s a problem. Randomness is exceedingly important to making computers and networks work.