The laptop would pay a 50% fee over the (declared value + shipping cost). Couriers will mostly deal with that on send, but if sent through regular mail you need to declare and pay before you get it.

If you didn't include your tax number as part of the address (doesn't matter in which field), there's a non-zero chance that the package will be lost, held indefinitely or returned to sender.

It's great that there are people willing to help even in these conditions.

I don't remember what the 27B was, I tried a 27B with different quantization at some point for that one, but I settled on the 31B.

Investments shift to things whose tax value updates slowly, for example property which typically adjusted more slowly than other financial assets. This tends to rise property prices and concentrate ownership.

It causes other distortions in allocation depending on the tax details, but wealthy people tend to adjust more aggressively to changing conditions.

I like text selection exactly how it is. I want precise controls.

It's fine for a touch interface like a phone, but on a computer I expect precision. As much as I can get.

I select long text when I'm reading, I use it for focus and to keep track where I was if I need to switch to some other task in the middle.

In general, excessive customization is a net negative if it breaks expectations.

Perhaps a per node hash of known recent routes to avoid flooding every single time and using flooding as a backup.

I don't think it will matter much. They live off the exclusives.

If the rng is not customized it will use:

    const rnds8 = new Uint8Array(16);
    export default function rng() {
        return crypto.getRandomValues(rnds8);
    }

Unless an intermediate node lies and doesn't decrement and retransmits anyway.

    - current CapEx will make the production side increase capacity
    - advances in TPUs, NPUs, open weight and quantization will keep going at a rapid pace
    - when the spending slows/stops, hardware prices will drop, hard
    - most AI workloads will move to the edge (except frontier models) because the hardware is cheaper than a subscription

For example, most of my AI use lately has been running Qwen3.6-35B-A3B-UD-Q8_K_XL on a 64GB MacBook Pro with an M3 Max. It runs at ~57 tokens/s and it's mostly fine.

I do use the frontier models a bit, but only when the task is too complex for the local model.

Basic crap, like analyzing an existing codebase and bouncing ideas, making small changes, the local model is enough.

I mean these models are inherently probabilistic.

If you run enough samples you'll get results matching the learned probability distribution, the more you sample the higher the chances that you'll land on an unlikely response.

You meet a vegan crossfitter that uses Arch, what does it tell you about first?

That reliance on third-party AI is a huge risk, just saying.

I would love to see an instruction set reference for one of these, all you have is hardware architectural diagrams or high level APIs.