somewhereoutth

Humans have a 'world model' beyond the syntax - for code, an idea of what the code should do and how it does it. Of course, some humans are better than others at this, they are recognized as good programmers.

It could also be because we don't want a theocratic regime with nuclear weapons and the delivery systems to put them on European capitals.

It could also be that the whole region is simply tired of their bullshit, and would like to normalize their relations with Israel and generally get on with life without nonsense like Hamas and Hezbollah interfering. Note how Lebanon is seeing this as an opportunity to get rid of Hezbollah once and for all.

Probably a continuation of the 'mowing the lawn' strategy (as used against the Palestinians). Every now and again use massive military force to set back Iran's capabilities, time and effort they spend rebuilding is time and effort not spent causing problems elsewhere.

> Code never was and never should have been the product in and of itself

Except that the code completely and precisely defines the actual product. Bad code => bad product.

> code should be sacred and revered by itself

As a production of the hand and mind, code should be revered - if only as the mark of the human or groups of humans that made it.

> the wrong group of people

The group of people who care deeply about the world around them.

I'd suggest that a measure like 'density[1]/parameter' as you put it will asymptotically rise to a hard theoretical limit (that probably isn't much higher than what we have already). So quite unlike Moore's Law.

> Instead of only allowing programs that are proven to be typed correctly you might want to allow all programs that you can not proved to be wrong. Lean into gradual typing. Everything goes at first and the typing becomes as strict as the programmer decides based on how much type information they add.

Yes, this is the way. And if you ensure that the type system is never abused to control dispatch - i.e. can be fully erased at runtime - then a proposed language supplied with some basic types and an annotation syntax can be retrofitted with progressively tighter type-checking algorithms without having to rewire that underlying language every time.

Developers could even choose whether they wanted super-safe checking (that precludes many legal forms), or more lenient checking (that allows many illegal forms), all in the same language!

I would add that human interfaces need to be tested by humans (preferably not your paying customers), machine interfaces by machines.

Uber's genius was getting their workers (sorry, 'contractors') to carry the capital costs of providing the fleet of vehicles they use.

> What began as a jagged, precise Romanesque structure of stone is eroded into a polished, Baroque plastic shell

Not to detract from the overall message, but I think the author doesn't really understand Romanesque and Baroque.

(as an aside, I'd most likely associate Post-Modernism as an architectural style with the output of LLMs - bland, regurgitative, and somewhat incongruous)

Agreed. Many of the suggestions are pretty much code it yourself, but without actually tapping the individual keys.

Furthermore, and more generally, one of the great things about (traditional) coding is that it allows 'thinking through making' - by building something you learn more about the problem and thus how best to solve it. Code generation just leaves you with reviewing, which is less powerful in this way I believe. See also 'thinking through writing [prose]'.

No.

When there is low prevalence of a condition, but a non-zero false positive rate of a test, the false positives generated by universal testing can in fact be a net dis-benefit (worry, invasive further procedures, etc) to the patient population as a whole, regardless of cost. This is a well understood statistical phenomenon, and is carefully considered by healthcare systems when advising on testing.

"Hell is other people's code"

Not sure why we'd want a tool that generates so much of this for us.

Infinite scrolling combined with the algorithmic feed is the real nasty.

Feeds should be heavily regulated, effectively they are a (personalized!) broadcast, and maybe the same strictures should apply. Definitely they should be transparent (e.g. chronological from subscribed topics), and things like veering more extreme in order to drive engagement should be outlawed.

> based on multiple public statements from people like Dario

> clear picture of the finances of these companies

Since they are not publicly traded companies, presumably there is no legal duty for the officers to be clear or even honest about these numbers?

But even assuming good faith, my understanding is that the scale of the current build out is so huge that revenues would need to exceed the size of many entire industries to have a chance to turn a profit.

Sorry - if a model costs (say) 20B to train, lasts 12 months before it becomes obsolete, generates 2B/month revenue, but with 1B/month inference costs, then it has lost 8B.

Or are you suggesting that in fact each model comes out ahead over its lifespan, and all this extra cash is needed because the next model is so much more costly to train that it is sucking up all the profits from the current, but this is ok because revenue is expected to also scale?

Apple was built on a product manager shouting at engineers until they got it right.

"we don't want your electric car, we want your car, but electric!"

Indeed. And this burden of a top heavy population pyramid is a major reason for not having [more] kids - a vicious circle, which, if left to continue, will result in humanity simply evaporating.

A pretty depressing place, with whole towns and cities abandoned, as the dwindling population huddles closer together. Not just geography though, there would also be a retreat in the arts, sciences, etc as there are simply not enough people to maintain let alone advance these endeavours. Life would be about eking the last out of what was left over from the 'glory days', a sort of slow motion apocalypse.

However a lot of the modern world is carried by 'pride of workmanship' - and not just by those who 'make'. It's an extension of the 'planting trees' parable, to care about things even though you are not immediately (or ever) rewarded.

However if the 'non determinism and inaccuracy' of LLMs is more pathogenic than that of humans, then we have a problem. Pathogenesis is inherently a system level effect, so it may take a little time (and money!) to become evident.

specifically to keep up with the Soviet Union. Sputnik 'concentrated a lot of minds' as they say.

I suspect a lot of this is due to large amounts of liquidity sloshing around looking for returns. We are still dealing with the consequences of the ZIRP (Zero Interest Rate Policy) and QE (Quantitative Easing) where money to support the economy through the Great Financial Crisis and Covid was largely funneled in to the top, causing the 'everything bubble'. The rich got (a lot) richer, and now have to find something to do with that wealth. The immeasurable returns promised by LLMs (in return for biblical amounts of investment) fits that bill very well.

The cowling is its own structural support, so needs to be strong all around, otherwise it would fail on the other side and you'd get blade+cowling approaching the fuselage at high velocity.

I think it is the stators fixed to the engine nacelle, judging by the article image.

In other news: https://fortune.com/2026/01/29/100-percent-of-code-at-anthro...

'ability to reason' implies that LLMs are building a semantic model from their training data, whereas the simplest explanation for their behavior is that they are building a syntactic model (see Plato's Cave). Thus without new training they cannot 'learn', RAG or no RAG.

without training new models, existing models will become more and more out of date, until they are no longer useful - regardless of how cheap inference is. Training new models is part of the cost basis, and can't be hand waved away.

LLMs will stop being trained, as that enormous upfront investment will have been found to not produce the required return. People will continue to use the existing models for inference, not least as the (now bankrupt) LLM labs attempt to squeeze the last juice out of their remaining assets (trained LLMs). However these models will become more and more outdated, less and less useful, until they are not worth the electricity to do the inference anymore. Thus it will end.

Features : Quality : Timeline

Choose 2. For example a large feature set can be made quickly, but it will be of poor quality.

Note that cost is somewhat orthogonal, throwing money at a problem does not necessarily improve the tradeoff, indeed sometimes it can make things worse.

> Investors were briefly spooked last July when an M.I.T. study suggested that almost none of this is useful to businesses. Corporations had poured tens of billions of dollars into A.I., yet only one in 20 projects had succeeded, the study reported. But a Wharton study in October delivered the opposite verdict. After interviewing 801 leaders at U.S. companies, Wharton concluded that three-quarters of the businesses were getting a positive return on their A.I. investments.

MIT actually carried out a study. Wharton just asked some execs, who of course parroted the party line. Winter is coming.