They built a custom engine for a specific model and it went 30% faster. That's great, but it's inflexible and of limited use.

Of course these days, coding agents make designing and writing new code very cheap. With the right tools and toolchains it shouldn't be hard to automate the process of building a custom engine for each new model that comes along. The cost of doing this would be small compared to the cost of training the model in the first place, so really we should just be doing this as a matter of course. For a 30% speed-up it would certainly worth the effort of building those toolchains.

Right, because there was no value to the US in having $19.6 billion of inbound investment into US infrastructure, or in having another large scale source of energy at a time of rising energy demand. And there as clearly no value to the US in leaving that $1.2 billion in the treasury. We're clearly better off without it.

Can I ask how you define "being political activists"? Does it cover simply pointing out that statements by some politicians are at odds with the evidence? Do you consider saying "The available data consistently indicated that climate change is the result of human activity." or "Vaccines do save lives; here are the studies which show it." to be activism?

I submit that the perception that scientists are being political activists has very little to do with what the scientists are doing and a great to do with politicians who don't like the science constantly complaining that scientists are activists and radicals. I'd be happy to run a study to try to measure that, but I suspect that it wouldn't sway the opinions of the people who wrong.

Did you look at the methodology that they used? Did you look at the way that they compared the pedestrian injury rates both over time and between different types of vehicles of the same age? Perhaps it's a tall order, but did you even read the article?

Imagine creating a model and now you have to wait for a 5 year-old to review your model for it to get released. If it doesn't say that that their favourite things are the best things ever then it fails the review.

Essentially all of the major scientific advancements that have led to the US's prominence in science in the last 80 years have been the direct result of government funding of the primary research. Industry and private enterprise are great at taking the results of primary research from the lab to the market, but the vast majority of investors have neither the foresight nor the patience to support the fundamental breakthroughs. Since the second world war the US government has been the largest funder of that, and the US has benefitted enormously for decades as a result. Now China is set to become the largest funder of primary research. You don't need to be a Nobel prize winning scientist to work out what is going to happen as a result.

They have indicated that their mysterious Mythos model is a lot larger, and will likely cost a lot more to run. Opus 4.7 is an incremental improvement on Opus 4.6, at the same price. That seems like a reasonable stop-gap until they have built enough power stations and bought enough GPUs to make Mythos available to more than a handful of users.

While I won't argue with speed limits often being arbitrary, the problem of fines (and civil forfeitures) being used to generate revenue is largely an American one. In most places fine revenue either goes to the central government's general fund or is ring-fenced for specific uses that are not the under the control of the people enforcing the laws. In the USA most states let the police agency that levies a fine or civil forfeiture keep some or all of the money. This leads directly to abuse. Enforcement is always subjective and selective; giving the police a personal financial incentive to be heavy-handed is bound to lead that way. Unfortunately there is a notable contingent in the US that seems to think that "law and order"(TM) is perfectly even-handed and should be vigorous, and also want the funding for government to come from someone other than them, so this sort of arrangement persists even though it's patently obvious that it leads to corruption.

The amount of precipitation wasn't the issue this year, it was the high temperatures in the mountain West. From the OP:

It's the snowpack that feeds the rivers through the dryer months. Winter rain melts what snow is there and it all runs down hill quickly (literally and metaphorically). There's only so much space in the reservoirs and the rest is back in the ocean by April or May. This summer is going to suck for farmers, or anyone else who needs a steady water supply.

Maybe, but that's also what people said about both Google and Facebook. The cost of delivering the service will go down over time. The cost of customer acquisition will go up as incumbent players solidify their positions. A loss leader now is arguably well worth it for the future. Add to that the fact that the vast majority of the investments being made are going into hardware that is increasingly hard to come by and the risk to investors is less than they first appear, while the upside is potentially huge.

There is a lot of hype in the AI space at the moment, and the valuations across the market as a whole are unsustainable, but Anthropic is one of the companies that seem more likely to come out well in the long run.

I grew up in a house that was more than 300 years old. There are plenty of churches in Europe that have crypts which were sealed 1,000 years ago and are still in fine shape. There are large scale human constructions such as the Egyptian and Mayan pyramids and the Great Wall of China that are over 2,000 years old. Construction to last 300 years is not that hard.

Probably more importantly, 300 years is short enough that people might actually remember why they are not supposed to go "in there". I suspect that one of the biggest challenges with burying waste for 10,000 years would people thinking that it looks "interesting" after 1,000 years and digging it back up.

No, sodium is much cheaper than lithium in the form that is needed to make batteries; recent commodity prices for NaOH have been 10x to 20x cheaper than bulk LiOH, although this isn't all the cost. The technology development is now to a state where complete sodium batteries are cheaper the lithium ones and has been for a year or so, and the technology is improving fast.

You are correct that the energy density is indeed worse, but that gap has also been closing in recent years. Modern Na batteries have better energy density than the Li batteries in cars from six or seven years ago. If you're trying to build a lot of cheap electric cars then the lower price is very likely more important than the cars being somewhat heavier and thus a bit slower.

They are not the only people to make a profit from AI!

Surly everyone knows that Belgium is just a little bigger than Lesotho but just a little smaller than Moldova, but for the geographically challenged, that's a somewhat bigger than Hawaii but smaller than Maryland.

It was the size of Luxembourg after the last downsizing. It's now the size of Andorra, but there's another set of cuts coming and it will end up a little smaller than Monaco.