ben_w

It's not "obviously logical", it's a pattern which we mimic to avoid mockery.

example For, semi-randomise I word order can this like, Yoda worse than, and be understood.

is no problem for a machine that takes context and probability into account when translating words to the underlying grammar structure.

We had to invent Transformers to be able to do that with reliability anything close to being worth caring about. Transformers have to learn from examples, not be pre-programmed.

I've found it's had unacceptable scroll performance in long contexts for a while now.

In this case, it takes me 12 seconds to see content when first opening the link, and about 18 to re-render content when scrolling.

If natural language was structured in a logically computable way, we'd have had interesting chatbots by the late 80s, basically as soon as a dictionary fit in local RAM, and for the same reason we got compilers.

Da hole raisin y nat-lang be v. hard is dat i kan rite lik dis an it be cool 4 native engrish speekrs 2 unerstand. LLMs are of course fine with this sentence in exactly the way that Zork's engine couldn't be.

ChatGPT repeatedly asserting that 0/x=0 is division by zero and therefore no value of x satisfies the equation.

Aye.

A decade or so ago I wondered if the reason maths was hard was the names being optimised for writing by hand. Everything's single letters if they can get away with it, so when mathematicians run out of Latin alphabet, they use Greek, bold, etc.

Even integration's ∫ is a fancy elongated s.

CS version would be e.g. integral(function=some_named_function, from=a, to=b, with_respect_to=argument_of_function), which may be longer, but is less opaque, especially when you get in so deep there's 3 other people in the world who've looked into this specific problem and you had to invent your own operations.

But that's all an outsider's perspective. I stopped with two A-levels in maths and further maths.

Everything has to compete with everything, but:

(1) agriculture is both the biggest user and not done in a way that comes even close to maximising land efficiency; nor is there any reason it should, given there's not much pressure on agricultural land and the UK has easy access to cheap food imports.

(Even if there was pressure on farmland due to a loss of trade, it would be access to fertiliser that would cause more long-term issues than the land itself; phosphorus shortages would reduce carrying capacity in the range of 9-39 million depending who I ask).

(2) even with the UK's right to roam legislation, leisure use of most land is close to a rounding error. I mean, think about it for a moment: unless you live somewhere like Cambridge where the cows roam in the middle of the city, how often do you go to the middle of a cow pasture for a picnic? Ever tried taking a shortcut through a field of sunflowers (I did as a teenager, I don't recommend it)?

That said, on (2), UK land dedicated specifically to golf courses is close to (a little smaller than) UK land dedicated to housing: https://www.bbc.co.uk/news/uk-41901297

Both statements are true without contradiction.

A specific model develops as it passes through training.

AI labs change the architecture between models to allow them to surpass the previous models' best scores.

It's also going to keep being a hard sell to say "stop anthropomorphizing LLMs" when the models anthropomorphise themselves.

But besides that: Yeah, sure, they're not human, they're a cargo-cult mimicry of by and of minds that popped out of evolution doing gradient decent on intergenerational survivability. So what? Still interesting when the result of such cargo-culting incidentally echoes what our natural-selection-not-engineered moist electrochemistry happens to do.

History is going to look back at this time and how we let such foolishly inconsistent people make such grand choices for everyone poorly.

Assuming we have a future history. We've already got "history slop" with AI rewriting the past by their incompetence.

Given they're "such foolishly inconsistent people", would you rather they err on the side of caution like this? Or the side of boldness, like Musk has been doing with FSD/Autopilot or Grok porn, all of which he's getting in legal trouble over?

I distrust Musk and Zuckerberg (to put it mildly), so it's fair if you say you don't believe anyone's public statements; but I also hang out with some of the researchers on this, and a fear of e.g. ending up with something as criminally unhinged in cyber-work as Grok was with porn is the least of their worries. Plenty of them also fear a corporation centralising power with such tools (such power is Musk's entire sales pitch for why line go up in future).

I don't think you can 100% ensure your queries have absolutely no biology and cyber keywords inside.

Obviously. Almost everything is a precursor to something dangerous, to the extent that if some model isn't aware of the risk it will wander into it blindly, e.g. suggesting leaving raw garlic and olive oil alone for a week without awareness this will likely breed botulism bacteria.

This makes models like Fable 5 impossible to use in any serious agentic task, because you can't even guarantee the model, which is a basic thing you need to build on.

This is binary thinking: "100% ensure", "impossible to use", "can't even guarantee the model".

Outside computers, most work is not binary, it's probability, e.g. "this skyscraper will probably survive being hit by an aircraft; oh we didn't mean a 747 we meant a small Cessna, but what's the chances of a 747 crashing into it soon after takeoff?".

Fable being too cautious for its own good (especially since the other models were not) is a fair criticism, but this isn't a binary question.

"thanks, you too"

:P

That's the neat thing, with LLMs we can fully automate this discussion and have it at thousands of tokens per second and millions of tokens per dollar.

:P

Let they who has never ='d when they meant to == cast the first argument.

(There's a reason why we benefit from code review).

I'm not sure exactly which enterprises you have in mind, but sure: quantities which can be expressed as "a year's worth fits on my desk" should not be described as "staggering", and 10 PB of hard drives will (just about) fit on my desk.

The LHC, on the other hand, that generates a petabyte a second and has to throw most of it away for obvious reasons:

https://www.itnews.com.au/news/computing-for-the-large-hadro...

Muskiest

I believe they mean Claude does the work directly, rather than writing software to do it.

"thanks, you too"

Yup. Been feeling the same for several years now, on all points.

The performance is bad; the developer experience is bad (it's a moving target, not a stable point); it was done to chase a trend (reactive) and to boost other Apple platforms (macOS, watchOS, tvOS, and now visionOS) rather than because it was the right technical solution; the UI and UX it generates is poor; and Apple (but not only Apple) seems to have stopped caring about quality over the last few years.

(On that last point, I decided to skip macOS 26 because of all the issues, but here's what I faced with their -2025 offerings: https://benwheatley.github.io/blog/2025/06/19-15.56.44.html)

If I were to make YouTube videos of the camera-to-face variety, I would cloak my face and voice both to avoid stalkers and to avoid deepfakes of each.

But against the flood of videos even before all the AI-generated content churned out by people who don't care about quality, I think I will instead just not make any.

Likewise.

I think it's a real pity that of the "four boxes of liberty", Musk has interfered with the first three: soap, Twitter; ballot, the lawsuits about million dollar rewards for voting; jury, selecting his jurisdiction, dismantling enforcement institutions via DOGE.

On the plus side, it's only "interfered with" rather than "completely eliminated".

Still, glad I have an ocean between me and the USA.

It's much easier to focus on yesterday's threats than tomorrow's. A drunk looking under a lampost because that's where the light is, even though the keys were lost somewhere else.

That said, the AI companies are one of the few places where they take future concerns so seriously, that they entertain concerns most people observing them think are head-in-the-clouds-sci-fi-levels-of-delusional, e.g. "what goes wrong if it works?"

This does not make them correct about the threats of tomorrow. Prediction is hard, especially about the future.

Singapore, I'm told, from someone who lives there. I don't really follow their politics so take this with a pinch of salt appropriate for hearing things on the grapevine.

France also has four times the land area compared to the UK.

This is not by itself of great importance.

There's several different meanings of "built on", but even with the most expansive definition for this purpose, the UK could double how much is "built on" at a cost of around a tenth of what isn't: https://fullfact.org/economy/has-92-country-not-been-built/

There's other reasons to say "perhaps not a good idea" to a higher population (food security, energy security), but the land is there for more (or simply bigger) housing.

Baumol effect is my best guess. Nursing tracks the cost of labour, the tech is (so far) making a small fraction of the worldload a lot more efficient, but the bulk of the workload is not (yet) amenable to automation.

That said, I do hear people sometimes complaining that the administrators were fired to save money, and this just put all that administrative burden onto the doctors and nurses who weren't trained in it and didn't go into medicine to perform; if that's true, tech may be able to assist with reducing that burden?

The problem with changing the 'free at the point of use' principle is that under any likely replacement it would continue to be free for expectant mothers, children, the disabled, and the elderly - precisely the people who consume the largest proportion of healthcare resources.

The disabled were scapegoated for much of my teens, I wouldn't be surprised if that continues; the elderly… I won't be surprised if someone realises an old person who needs treatment to live, doesn't get it, dies, isn't going to vote against the party responsible for that in the next election the way a pensioner whose income doesn't rise with the cost of living will be able to.

Mothers and children are still going to be important to look after. But with ever-decreasing fertility rates, I wonder if this has become much cheaper per capita already, and the "true" (steady-state) cost for this part of the NHS is therefore hidden, just as old-age care costs were hidden when the population pyramid was the old shape?

I'm not sure what you're arguing here.

Upthread, the point about it not mattering what China stole but what they can do? That "Made in China" has gone from a sign of low quality to being the default as it became the factory of the world? Yes, China's winning and the sooner the rest of us wake up and smell the tea the better. They learned lessons from how my ancestors were able to push them around despite coming from a small, damp, sheep-filled rock in the Atlantic, and don't want that to happen again; for most of "the west", being on the receiving end of such humiliation is a historical footnote if it's in our own history books (or living history) at all*, and most of the exceptions are former-Soviet-bloc/Warsaw Pact.

But specifically to the point about the USSR claiming all (good) things for itself, as parodied in this manner by Star Trek? AFAICT, that was just plain soviet propaganda.

* What happened in Africa and India/Pakistan, however, is recent enough to still be in living memory. Israel likewise, though now this is on the edge of living memory for the things which led to its reincarnation.

And while the Irish definitely still retell history lessons about the British and the Famine, I'm not at all sure if the stuff in NI in my lifetime counts as "humiliation" given I'm British and our newspapers just said "terrorist" about everything that came from there.

This is only relevant to the point that most of us are deeply oblivious to how bad things can get when someone else is the boss and we don't get to vote in their elections.

Probably sooner, IMO. Absent radical life extension, even 200 years from now there'll be nobody's left who remembers the time before deepfakes, and history will be brought into doubt.

Though bluntly we've always had nationalistic narratives contaminating what's "approved" in school, both with literature and history. Stephen King could be anything from mandatory to forbidden for any reason at the same time in different places, even on that kind of timescale. The USA may cease to exist as a singular coherent entity, and it may be two or more successor states in parallel who have a culture war with each other over if Stephen King was real or not.

Disagree, I think it's analogous to how we today use "fire" to describe the action of loosing an arrow or whatever the technical term is for a trebuchet, even though neither has gunpowder.

Mainly βωμολόχος, I think? But also ἀλαζών in some episodes?

I think it may be cheaper to use Starship as a point-to-point delivery system to put solar panels and compute modules in a desert, than to put the same compute performance into space.

Mainly depends on how much of a mess Starship makes taking off and landing.

and they're probably going to be the only company with a fully re-usable launch vehicle. Nobody else except maybe for some chinese companies and kind of RocketLab is even close. They basically have a monopoly...

Back in 2018 when dearMoon was announced, I was expecting this to be true for a while, a temporary monopoly while others caught up; but dearMoon was due to launch in 2023, got cancelled in 2024, and SpaceX have so little confidence in the vehicle dearMoon was supposed to use that it still hasn't had a fully circularised orbit so far, by mid 2026.

This means Starship has been slower to develop than the Saturn V and the Space Shuttle, with many of the same problems as the Space Shuttle and the Soviet N1. Rather loses its shine when that's the comparison.

Honestly, I think people are asleep at the wheel with the valuation and shorting it seems stupid. They have the star shield thing, and hypothetically point-to-point QRF systems for governments with star ship. Then there's space data centers.

If nobody else had a chance of catching up in twenty years[0], and if the world trusted Musk[1], then Starlink would be worth something like 100-200 billion in today's money. Realistically, it's probably more likely worth $50-100 bn.

Starshield and QRF is ??? money, because while the US government is very good at spending taxpayer money on military stuff, Musk's personally made a lot of political enemies and may just not survive a Democratic congress or President.

The space data centers… I really need to get back to editing my draft blog post, I've got a whole bunch of open threads I'm pulling on and around 8k words on why all possible variations of it are a bad idea. At presented scale, it's about the same difficulty as making an orbital ring[2], which would be a better use of resources, though still has the geopolitical trust deficit working against it.

[0] This is the relevant time horizon for valuations

[1] Highly divisive; while this doesn't matter in all markets, it matters more in the markets with more money to spend on Starlink.

[2] Contiguous structure with two parts, one stationary with respect to the ground, the other slightly above orbital velocity to provide additional centrifugal force to keep up the stationary part. Makes space elevators much easier, because you only need 100-200 km rather than 36,000 km.