eigenspace

1 in 200 apartments in the building? You realize that the USA is one of the only car markets that has closed itself off to chinese cars, right?

BYD is the number 1 EV brand in the world currently by sales.

Renewable-heavy grids like Germany use overlapping layers of protection for this sort of thing.

Tier 1 is flywheels because they have the fastest response, and then response isn't regulated by software but instead is purely mechanical

Tier 2 is batteries, they can be made to respond to current or frequency faults in a response time on the order of miliseconds, and the flywheels give you plenty more time than that to make grid-wide orders for the batteries to respond

Tier 3 is gas peaker plants. The batteries are able to buy enough time that if there's a serious fault that could last multiple hours, you're able to spin up (or down) a gas peaker plant to adjust the grid frequency / current.

Intuition is much more adjustable than youre giving it credit for.

One can build new intuitions.

What on earth makes you think that?

The learning rate for a solar panel may be higher than a wind turbine, but when you include installation costs, it's no longer the case because the solar panels have become so cheap that the cost is installation-dominated, and installation learns slow.

Over the last 5-10 years, wind has been learning faster than solar.

So there's the hobbyist side of it, but I guess it's not totally clear why that matters to the people talking about the world's power mix.

But as I write that, I guess it becomes more clear to me that a personal interest in installing the technology makes one pay more attention to it elsewhere, and promote more conversations about it that get others involved.

Wind power scales with the turbine area squared, and the wind velocity cubed. That means there's massive incentive to make bigger and taller turbines (windspeeds are higher the further away from the ground you are).

For that reason, there is very little reason to build small wind turbines, even if there wasn't a down-stream loss of power from putting too many close together.

I find it continually strange how much more attention solar gets than wind.

Wind power has been undergoing a very similar rapid fall in cost, and rapid rise in installations, and is a fantastic natural resource (and also one that complements solar almost perfectly!), but it always feels like solar gets 10x the attention and conversation that wind gets.

That's a privacy claim which is irrelevant to the question of sovereignty. Even with zero-knowledge processing, they can still just close your account and stop providing any service after a phone-call from the US president.

Vital government functions cannot be allowed to run on this type of infrastructure, it's ridiculous.

It's especially funny that Microsoft has been swearing up and down that they can offer "sovereign" infrastructure in Europe, and yet the moment Trump comes calling they trip over themselves to implement his sanctions against the ICC.

While I agree to a certain extent, Americans aren't exactly helping the case against them.

two thirds of you didn't think it was worth voting against.

The Americans who saw this coming and stayed home are just as complicit.

Why mention nuclear but not wind? Wind has nearly as low an LCOE as solar, and is still dropping too, and wind power is anti-correlated with solar power, meaning we get more wind at night and in the winter.

Even if the number of gas turbines will increase (it's not increasing nearly as much as you'd think if you just look at the number being built, because old ones are also being decomissioned), the amount of gas being burnt on an annual level is going to be going down. These gas plants are for bridging gaps left open by wind and solar which cannot yet be covered by batteries.

If you believe in the transition to renewables, Germany fundamentally does still need gas turbines to make that happen, more than we currently have.

> You do realize that China is not the only place making open weights models, right?

It's the only place making open weight models close to competitive with American closed-weight models, and the only open-weight models that beat Mistral's models.

> I don't think they're justified now, certainly their results don't justify it. Mistral is mid at best.

Their investors and customers disagree.

> Except they don't, they are dependent on GPUs from nVidia and AMD—both American companies.

That's a pre-existing dependency that nobody in Europe is happy about. The goal with Mistral is to avoid there being yet another new layer of outside dependency. Mistral isn't a GPU company and isn't trying to be.

However, if someone in Europe was able to build a GPU that's as close to Nvidia or AMD as Mistral is to Anthropic or OpenAI, that company would also be seeing a lot of keen interest and likely piles of money.

What happens if Mistral stopped trying to make foundation models, and then China stopped releasing their open weight models?

It's core to their whole value proposition that they are providing sovereign AI capabilities in Europe. If they're just repackaging Chinese models they don't really provide that sovereignty, and they'd be unable to justify their valuation or investments.

I suspect the reasoning trace API is just bugged.

Thar looks like a reasoning trace summary.

I didn't say they did (though granted, I believe it's the case). What matters here is that countries without their own AI labs have multiple potential partners to choose from.

Having a third option significantly lowers the leverage that both China and the USA have over other countries.

That cluster is literally orders of magnitude smaller than the compute pools used by Anthropic or OpenAI.

"We investigated ourselves and determined that we are not to blame"

I use Opus 5.5 daily for my job. I am aware (and in awe of) it's capabilities.

Look at the context in which I used that term 'good enough'.

What i was saying is that there are tasks for which a dumber model can be good enough, and for organizations with sovereignty/ privacy concerns, those concerns can be strong enough to incentivize the use of a dumber model.

Then you dont know people working on highly sensitive info with stringent privancy concerns.

Yes we absolutely can compete. That's such a dumb defeatist attitude.

And I dont mean actual trade stopping in the case of AI. The CCP could ban the release of Open Weight models tomorrow without running afoul of a single trade treaty

Regardless though, their reason to exist is not to advance the state of the art.

Their reason to exist is to ensure the sovereignty of France.

Obviously they would do their job better if they were advancing the state of the art, but it's not like it's pointless if they arent the absolute best.

Using Chinese models is a bit like using Chinese solar/batteries instead of American gas.

Yes, if the Chinese stop trading, you can still use the existing solar panels (unlike the gas which you literally set on fire), but it's nonetheless a major vulnerability to not have any local know-how in creating important infrastructure.

Giving up all AI know-how and expertise to China just because they currently share their models would be a generational mistake.

Europe already got burned hard by this sort of thing too recently, and is now very sensitive to strateigic depencencies, and is working hard to lift them where possible.

The big lab revenue may not be catchable, but im not sure it needs to be.

If they can carve out a niche of industrial and governmental partners who rely on them for sovereignty reasons, it may be enough.

I agree that Opus and GPT are almsot surely better, but so many real users are nervous enough about giving Anthropic and OpenAI access to all of their internal information that they may be willing to stomach worse models if it gives them more security.

The real question is if this model is good enough that it can still accelerate work, and not be a hindrance to real work like older Mistral models often were.

If they can do that, they'll have customers.

Has the French military advanced the state of the art on any dimension in the last hundred years?

And if not, why do they exist?

Its quite interesting to see that at least the early days of AI so far have not been a winner-take-all runaway acceleration game where catchup is impossible.

I certainly wouldnt have predicted that 10 years ago.

Very glad to see Mistral still in the game even after some big stumbles with Large 3. I deeply hope that this model is 'good enough' that it becomes the European go-to, giving them the resources to keep the pace up.

I'm excited to try this out today.

To Europeans at least, this is a big deal, especially because of cybersecurity capabilities.

It is of vital strateigic importance for Europe (and really the rest of the world too) that there are non-American, non-Chinese options for AI.