whoisburbansky

It sounds crazy because it's not true; the Duo weighs 8.96oz, or 254g, a far cry from the MacBook weights.

The only (tiny) issue I've had with Tailscale on Kobo has been that the tailscale daemon prevents me from using the Kobo in Mass Storage Mode while it's active, so I have to disable/quit KOReader to be able to plug it in again, which is admittedly not frequently warranted anyways.

OTOH, koreader on the Scribe post jailbreak is easily the best large format PDF reading experience I've ever had.

Profits from the stake lower taxes that would otherwise be levied on you? Of course that’s moot if the deficit isn’t something being taken seriously.

No, of course, but is the primary focus of the bulk of the staff educational or patient care? Seems disingenuous to pretend it's the former just to make a point.

I mean, if you tack on a hospital to a university, the correct denominator to compare against is "patients served," not "students educated," at least for the portion of the headcount you're sticking in the numerator.

The key idea of Latent MHA is that "regular" multi-headed attention needs you to keep a bunch of giant key-value (KV) matrices around in memory to do inference. The "Latent" part just means that DeepSeek takes the `n` KV matrices in a given n-headed attention block and replaces them with a lower-rank approximation (think of this as compressing the matrices), so that they take up less VRAM in a GPU at the cost of a little extra compute and a little lost accuracy. So not caching, strictly speaking, but weight compression to trade compute off for better memory usage, which is good because the KV matrices are one of the more expensive part of this transformer architecture. MoE addresses the other expensive part (the fully-connected layers) by making it so only a subset of the fully-connected layers are active at any given forward pass.

Foundations don't change much with "the speed of development"

If the texts were incomprehensible to parents, how were they comprehensible to their intended recipients?

It plays out this way because anything anyone buys with a credit card, reward card or not, ends up costing 2-3% more than it would otherwise have, because of interchange fees. If you have a rewards card, the CC issuer turns around and gives you, say, half of that back (1% cashback on everything) and keeps the rest. It's kind of like a tax break that you only qualify for if your credit score is above a certain threshold, but you have to pay into regardless of income/credit score.

It depends on what your goal is; for some tasks it's possible that getting the AI to do it is best, but, e.g. the existence of auto-pilot doesn't mean that hobbyist pilots wouldn't benefit from/enjoy exercising the same skills manually.