AI produces output that is very convincing to a non-expert, and (dangerously), it's so good at looking like an expert, they might believe that it is an expert. But the moment you ask someone to use it for something they're an expert in themselves, the holes appear wide, consistent & obvious.

My favourite moment of seeing this in action was watching AI-worrier TV host/comedian Bill Maher. He has spent years talking about the dangers of AI taking everyone's jobs, destroying civilisation, ruining the economy, starting wars, "it's just getting better and better all the time", and so on. But one night he let slip a tell. "It's no good at writing jokes. Not yet, anyway". There you go, Bill... connect those dots...

There is real utility in it being a tool to help experts apply their expertise, as in this story where it speeds up some tasks to help the translator do part of the work, enhance their expertise, allow them to be more productive.

It's a better screwdriver, a better hammer, in the hands of somebody who knows what needs a screwdriver or a hammer. It doesn't replace them. It can't replace them. It's a tool that enhances the human, not an alternative.

I don't understand why this is not widely understood yet, but I'm sure it will in due course.

And I don't expect this to change. Even if the latest model scores 100% on every benchmark, all that really tells us is that it's now more productive/efficient than it was before at helping experts do that work, not that it can replace everyone in that category of work.

I bet there is a river somewhere near you. Explored all of it? What about a hill? Is there a road you've never driven down? It might have some stuff down it you've not seen before. Have you explored all the flora and fauna around you? Obviously you need to stay off private land, but I would be amazed if there is absolutely zero variation in any topology, geology, animal/plant life, or other factor within a 100 mile radius of you.

If that is the case, can you tell us where that is? I want to visit exactly once and never go there again. It sounds both magical and terrifying in one instance, and reminds me of a friend who drove down Route 66 and found the expansive empty plains "the most claustrophobic thing I've ever experienced in my life".

They also don't know much that you can probably make use of. They might think they do, and you might think they do, but they got there mostly through knowing how to talk to boards and investors, not by being able to engage deeply in expertise that is applicable to most people looking to make their way in the World - and if becoming a CEO of a major tech firm or President is the thing you need the help with, you probably know them or people like them already.

I've met quite a few famous people in tech over the years, particularly open source, and have had some short and some long conversations with many of them. I've found most people pretty approachable.

I also know through another side of my life quite a few people in the media and am an acquaintance of someone who is a household name in the UK. Through him, I've met famous sports people, writers, actors, etc., and through that and other networks I know people who have worked behind the scenes on major TV and theatre shows who have met hundreds of famous people.

The one thing that unifies all of them is obvious, but seemingly lost on a lot of people who "other" those whose names are known to them despite never meeting them: they're all just human.

They're not "other", they're us. Including everyone you see on TV, everyone you have read about in magazines, everyone you see on a stage.

They have to put up with being recognised and people dealing with them in strange ways (how would you really deal with a stranger asking for a selfie while you were eating dinner with your family in a restaurant?), but they still do all the things you and I do. As the old saying goes, they all have to put their trousers on one leg at a time in the morning.

I'd definitely encourage people to seek out experts (not just "famous people" unless those people are famous for expertise), and engage them as you'd want to be engaged about your expertise. You'll find most people will be approachable.

But emailing that specific list of people is unlikely going to get you much beyond a template reply from one of their army of assistants.

The content is great. The tool gives writers a low-friction mechanism to charge for premium content, and works on most people's devices. I would rather have read this on substack, than the author get frustrated at having to learn how to publish pages by hand and give up.

Maybe we just need a better alternative to substack if that's the problem.

JIT as an energy saver intuitively makes sense, and is probably the model most languages need to think about for "shipping to prod". I'm aware Python has started developing this, and given the install base, it's encouraging that results like this show it could have significant benefits for users.

I remember people could smoke on planes. On some airlines seat backs and bathrooms had cigarette ashtrays in them. Smoking was phased out between 1988 and 2000, with most airlines being smoke free in the mid-1990s.

But the ashtrays persisted well into the 2000s. Two reasons: they needed to refresh the cabins, which is on a longer maintenance cycle done every few years, and before that, they needed replacement seats and bathroom fittings without the ashtrays. That meant tests, regulatory approval, all sorts.

For ashtrays being removed.

Winglets are a similar story. They're an addition, but they needed test flying and type approval before they could be added to the maintenance cycle rotation and get added to aircraft.

This is a bigger change. Boeing and Airbus (and others), are going to need to design it, push it through CFD, build different variants, test fly them, get them through regulatory approval and then... well, existing aircraft are probably not going to get these. Too expensive, too hard.

What's going to make more sense is a new aircraft - even if it's a variant type like the 737-MAX or the A320-Neo or whatever - where they approve the type modification as a whole, but it's not a retrofit to an existing airframe, will help manufactures sell more aircraft, airlines don't need to ground existing fleet and over time the fuel efficiencies get involved.

Most people are not doing that though. There's probably a good reason, and it applies to other languages too.

I have seen some of the newer projects, and like I say, the Raspberry Pi stuff makes programmable computing accessible to a kid without much, I just don't think the bar overall is as low as it was for me.

And yeah, survivorship bias, and a weird population skew with me: I was literally the only kid in that computer room determined to learn how to code.

That book started a remarkable journey. By 1996 I was at University studying Software Engineering, already proficient in C. Ten years later I was running my own software consultancy. Ten years after that I had been CTO for three startups and moved to London.

I often like to haunt a bookshop or library, and check out the programming books there. No 11 year old would be able to get started the way I did today in that context. I love the Raspberry Pi project and its goals, it's the closest we have to that opportunity. I do - and will continue to - support it multiple ways, and hope others do too.

Honestly, without those introductory guides to coding, I don't know what would have happened to me, but the odds say, considering what happened to my classmates from that school, drug overdose or prison were on the cards.

Thanks Usborne. Thanks BASIC. Thanks to that computing teacher who had that idea.

The judge asked if he was interested.

Perhaps the judge asked this knowing that the circumstances showed this was a caring man who had the child’s best interests at heart and had demonstrated through actions - and described through testimony we have not heard - his feelings towards the child when finding him.

They did not just get given the child. There was still a process. They visited the child in care. They filled in paperwork. They were vetted. They were asked if they’d like to look after the child over Christmas - not forever, not straight away. The process took a little time, it just took a lot less time than if the child entered care and they had to find other adoptive parents.

The most important variable to identify in this situation is capacity to love and care for a vulnerable child. Financial stability and good character still need to be there - and it sounds like they were identified before the adoption was completed - but the head start was there.

He was not left in a bin or dumpster. This is not an ideal way to give up a baby for adoption, but don’t assume he was unwanted, or unloved: you don’t know - or seem able to imagine - the full story.

Perhaps some professional programmers know how to write secure software in a language with undefined behaviour. Maybe we should think about that more rather than just writing off an obviously huge success as a failure?

"All" a model is doing is predicting the next words, based on the statistical distribution of words it has seen similar to the ones read/produced so far.

We push a model towards a particular set of distributions through context. If I ask a model "What is the capital of France?", there is a non-zero chance it goes down the dad joke answer of "The letter F". The far more likely option is "Paris", because the joke appears much less often in training material, but if I wanted to be absolutely sure of getting a consistent geography answer I'd address that with additional context. We can add context via prompts, RAG, agents, skills and so on.

However, when training a model, we select the material. We could show it a lot more geography information (or dad jokes!), and skew the statistical distribution in the direction we wanted. We could also decide to design the system prompt towards the direction we prefer - which the user would interpret as "the model" - and so nudge the context model-wide. We can also construct the interaction to iterate on context with a specific framing and call it "reasoning".

In this specific example, you could therefore solve the problem by a) training skewed towards mathematical papers, which likely degrades performance in general and likely for the specific case too, b) train the user to provide better context/prompts for mathematical work, shifting the workload to them which feels very "a la 2024", c) publish agents and skills that are tailored to mathematics work (very "a la 2026"), d) tweak the system prompt for when the model is doing mathematics work, which the user would see as "the model" doing the change, but you and I might look under the hood and say that is in the harness or a specific type of prompt, or e) add "reasoning" execution that is set to focus on mathematical formatting, or f) a mixture of the above.

Right now we're probably looking at agents and skills. I think over time we're going to see smaller models targets towards domains with a mixture of all of it, where some of this sits at user configurable levels, and some is "baked in" via training, system prompts and execution modes, but from a user perspective it's all just "the model".

Some people might interpret this comment as political commentary, but it’s actually just the reality of what people are saying and doing.

There’s a lot of data to suggest that America’s recent policy of reducing its soft power around the World & decoupling itself from alignment with interests of allies is causing increased interest and prioritisation of sovereign capability across tech, defence, public health and policy programs.

This was a campaign strategy/promise for the US President. I’m not going to comment on whether it’s good for the US or for the allies, but I will note it could have been better anticipated by all: the only real surprise is the speed and depth.

It raises some interesting questions - it’s one thing to say you don’t want Microsoft or Starlink in your infra tech stack, or don’t want to use AWS or GCP, but where does the line stop? Does the UK get out of Trident? Does the UN General Assembly get out of New York? No idea, but the fact these are conversations probably happening right now is remarkable.

It's policy. It's official Whitehall policy.

As a department you can't hire programmers at £100k/year, because that pushes them way, way higher than civil service bands allow. But you can pay a "Systems Integrator" - a consultancy like Cap Gemini, Deloitte, Fujitsu - £600/day for the same programmer in the same seat. So, £100k/year = bad, £120k/year via an external consultancy = good.

Then we get into actually building and owning tech. Look at the history of GDS - they were empowered to pay half decent salaries and build and own things, but then had budgets slashed and programs cut. Why? Because we can "just buy it". Yes, you won't own the IP, it'll cost 4x as much, it'll take 3x-5x longer, but at least you won't have "inefficient civil service bloat" to have to manage.

This all started in the 1980s, and there are signs of it swinging back. I was at one department last year where they were telling me they're thinking about hiring actual engineers and embedding some devops stuff internally - absolutely jaw-droopingly revolutionary. Genuinely.

I scanned a couple of chapters and realised it likely wasn't LLM generated, it just needed an edit. The intro to C is a hard and weird intro, but then driver development in FreeBSD is hard and weird and people who aren't prepared to get through such intros probably aren't going to get through the rest of it.

Being the contrarian, I've started going through it. I was involved on the periphery of the FreeBSD project ~25 years ago, went to conferences, ran a BSDUG in my hometown, and so on. And I realised I've missed systems programming and FreeBSD itself a little, and in recent years became a little sentimental.

What I've discovered so far in the first few chapters:

1. I miss FreeBSD. And it's weird my muscle memory kicks in and am surprised in a lovely way to find familiar things like /etc/rc.conf work the way I remember them.

2. This is not AI slop. There are issues that I can blame on him not using the same platforms I am (if you're on Apple Silicon, just use UTM and the aarch64 ISO - don't use the VirtualBox config he suggests, as an early example), but as somebody who sees a lot of AI generated content in my day job - this isn't it

3. I have got excited about coding again for the first time in a while.

So, this is my hobby for a while. Go back to where I started, get into low-level systems programming again, I have some ideas on some hardware I want to help out on... it's different to a lot of what I've been working on for the last decade or so, but that excites me.

[0] https://news.ycombinator.com/item?id=47915632

And of course, English develops organically (unlike, say, French), allowing new words to emerge, and for old words to take on new meanings. I love it.

As an Englishman, I always find it interesting that there is this weird defined notion of "Englishness" in language, culture, whatever, when our entire history is one of mashing and remixing ideas over at least 2,000 years, and recent discoveries at Stonehenge push that back potentially by 3,000-5,000 years more.

I particularly like the irony of the far-right going on about English identity on a march in London before going to have a lager and chicken tikka masala before heading home to a bungalow and putting on their pyjamas... :)

I think the Scandinavian roots you talk about trace back to common Germanic roots perhaps, but also the Viking aspect will influence a lot. I think English has been "dipped into" by those roots a few times in history, as has Latin.

On the need to keep the etymology aligned in translation: I think this is a routine challenge of the translator's skill, and why so many people have different views of different translations of the same texts.

The Bible could easily be translated in many different ways, but the "King James" version is considered the standard within the Anglican churches in the UK (and seems to be the common root for US church bibles too), but a more modern translation would be possible, as would one that has a closer etymological meaning to the original sources.

It's all interpretative. If people are building entire belief systems and ways of life (and arguably, laws for society), around a translation, and getting it off in a few places, it's likely we're going to run into the same problems even more when translating Tolkien or an ancient poem...

This mechanism is stupid, corrupt and driving extreme behaviour.

But betting on war is not new, and I find it abhorrent that people will hand wave away the military-industrial complex as an essential component to modern civilisation while calling individuals making bets on war horrible names.

It's just scale and semantics, not core differences to my eye.

Why? Well... suicides. You're going to start seeing gambling-related suicides in the next couple of years, if you haven't already.

If you can learn anything from the UK, learn this:

- VIP schemes kill people

- Online casinos and video slots kill people faster than sports betting do

- The industry will always "self-regulate" to the point where its still harmful, but palatable enough that law makers will look the other way to keep the tax raised from it coming in

I have skin in the game (I write code that profitably trades on sports betting exchanges), and think it's possible to make gambling a healthy and fun thing to do, but the American market is ~5-10 years behind the UK market, and I can see it's going to get ugly before it gets better (not that the UK market is perfect yet, either).