Search results

Surely the results of putting your customer support behind real term translation would be unusually hopeless even as these things go?

They're convinced that scaling further up will do some dramatic things. Very far from guaranteed of course, although it did earlier. Besides being true believers, I guess they almost need that because of there is a hard limit to how good these models get with scale, you eventually (quite...

With Tawain under permanent threat? Yes, obviously. Even in general you really shouldn't have something so crucial to essentially the entire global economy reliant on just one or two places. Of course, tariffs are a spectacularly awful way to go about trying to make it happen.

Not that I wouldn't cheer that last outcome but surely they're producing at least those at Intel or some USA native foundry? I do think it's a bit unhealthy to concentrate something so critical to the world economy quite so heavily. How to solve it I don't know (obviously not like this!).

Surely node pricing decays over time?

It's all so cell line/type/.... specific too, which makes combining large amounts of data really rather problematic. I got to work on a big project that tried to pull data out from the biology literature - a huge mess in it's own right! - then combine that into models. We did fractionally...

There was a huge shift with Alpha zero and it's ancestors. Not that anyone could actually beat the engines before then, but you could still tell there was sometimes something subtly 'wrong' about their play. Not anymore. There is it course an immense amount of skill in working out how to play...

They've been using reinforcement learning for at least 50 years by now :) The non trivial trick is always getting data and a reward scheme. Data is fairly easy for an LLM, the clever/new bits here are the reward mechanisms. Some of which do seem genuinely clever.

I think it's a bit stronger than this - it's known that DNA doesn't remotely dominate our lives in health terms. Because you can run the numbers and show environmental/behavioural factors absolutely dominating on a population level. Much the same ones we've known about basically forever, but...

It obviously isn't helping but absent infant mortality, these sorts of really major effects on life expectancy are nearly always down to quite basic, mostly obvious, public health issues. They just have much more impact overall. (The recent trend in the USA is obviously badly downhill there!).

My goodness, it's hardly like we're that amazing in the UK! Especially for the poor. As well as the obvious suggestions to date, how bad is the opiod situation still? That always seemed to be a basically American issue. The 'amazing' response to Covid is also probably going to be having an...

Isn't most of the global economy going to be kind of all right? It isn't like everyone is putting up the shutters at once, it's just America. Yes, the American economy is huge but compared to 'rest of the world' it still isn't so much.

If they're actually going to finish this (slightly mad!) exercise then never mind trimming, they're going to be slicing out non trivial plot lines left right and center :) In a few series time they'll be at the bit where that could sensibly escalate to scrapping an entire book or two. There's...

You could likely get a lot of the same effect by just randomising all the names and verbs?

Depends on whether it replaces one worker, or maybe ten. Or even none if most of what your people are actually is understanding requirements, tracking down bugs etc. Most likely, and it's a sobering thought, this is still priced to run at a loss (!). The amount of computation being burned is...

Certainly a brilliant essay writer. I love this little, somewhat happier one: https://www.orwell.ru/library/reviews/vicar/english/e_vicar

And that principally because as a group they're a lot poorer. It genuinely is moderately surprising if it has reversed this time.

There is a difference. Ukraine is broadly fairly terrible for the Ukranians, an invasion of Tawain would - as things stand - essentially destroy huge swathes of the worlds economy. It would takes years to get fabs up to replace the ones lost. I suppose that's whether or not anyone contests the...

Well, yes, but they are producing new models very regularly right now. Not as explicit updates to the underlying data sure, but that must be a side effect? If they stop trying to scale things to get new performance then it'll either be semi regular retraining or just letting the models worth...

Well, if people want to query said model about any sort of real world events then someone is going to have to pay to keep producing new models with updated data sets on a very regular basis. Which probably makes a lot of the potential economic models even less likely to really work out.