coolspot

Venezuela just flew their fully armed F-16s over a US destroyer. F-35 is a right tool to detect/eliminate those from afar. It can do land targets as well, if needed.

Too high. I think NK transit system is incomparably safer and cleaner than BART.

Riding without a ticket? Jail.

Littering on the platform? Straight to jail, right away!

Doing any violent crime in NK transit? Believe it or not - death by firing squad.

Here is a quick overview of how the system works: https://youtu.be/eiyfwZVAzGw?si=CnOMa8F6NkiyhifE

People do that because VISA credit card brand is spelled all caps on its logo.

North Korea? If you think it is a good example of a low bar of transit quality/safety to meet, then you’re comically far off.

Note that Proton is sketchy about their code being open-source and available for anyone to review: https://news.ycombinator.com/item?id=44665398

That’s the case in California for autonomous vehicles (like Waymo). Tesla FSD is officially a driver’s assistance, not fully autonomous vehicle.

> may you sip Starbuck's venti-size hot chocolates for many years to come

You’re basically wishing diabetes for him.

1024x1024 costs $0.03

Very cool! I was surprised how fast it generated an image for me! For better US adoption, I recommend to get a +1 phone number instead of +44 .

Common folk assume that the space is super-cold, so it must be good for hot servers.

Doesn’t matter for training, as all GPUs are colocated in the same DC.

We wouldn’t have Eminem

P40 has memory bandwidth of 346GB/s which means it should be able to do around 14+ t/s running a 24 GB model+context.

ClickHouse is just awesome and underrated. I tried to use TimeScale DB for 1-minute stock prices, and it was super slow and inefficient. On ClickHouse, the same data works, and it is blazingly fast.

I don’t think end users can pick a license because you have pieces licensed under GPLv2 and AGPLv3 . They will have to obey by both, which is tricky if they are conflicting.

It does feel like that for some people (like myself). But it was fun and informative to wear it once for 10 days.

For that case, Starlink is $10/m for 10GB, $50/m for 50GB or $80/m for unlimited (fixed location).

How much of context window does it take?

40 t/s

10B * 2000 t/s = 20,000 GB/s memory bandwidth . Apple hardware can do 1k GB/s .

They have provided repro for the 41.8% result here: https://github.com/QwenLM/Qwen3/tree/main/eval

> Lumo’s code is open source, meaning anyone can see it’s secure and does what it claims to.

No link to source code in the article. GitHub search also doesn’t show any source code for Lumo.

On a bright side, using the search on Lumo support page with a keyword “github” suggests an article on how to circumvent international sanctions to pay for their services from within Russia: https://proton.me/support/pay-russia

> That would be enough to support a single user. If you want to host a service that provides this to 10k users in parallel your cost per user scales linearly with the GPU costs you posted.

No. Magic of batching allows you to handle multiple user requests in parallel using the same weights with little VRAM overhead per user.

> Tesla has had FSD for years, and has an incident rate half of what Waymo has.

My research shows that you can arrive at these numbers only if you compare Tesla’s FSD incidents that are only reported if airbags are deployed, to all Waymo incidents, which are reported even when only a minor damage is occurred.

https://grok.com/share/bGVnYWN5_ae23a802-40b4-4545-a3a1-abb1...

Most of the defense tech is not using bleeding edge N2-N7 nodes.

As long as your page doesn’t wait for that image, your page is going to be shown faster if it is 14kb.

The Open-WebUI has a built-in vector database and document storage.

Github Copilot just added that about a week ago.

I think it is opposite. Flying into a mountain & etc would require one pilot to somehow incapacitate another pilot. Cutting fuel off, if done on takeoff, is not recoverable (engines can’t relight and spin up quickly enough).