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.
Hospitals attached to universities aren't in general "tacked on" but are a part of the educational environment. They exist not only to serve patients but to educate students.
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.
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.
Somebody else pointed out that it's likely just an alert, not a hard limit, which checks out given Firebase documentation (https://firebase.google.com/docs/projects/billing/avoid-surp...), which has no mention of hard limits and explicitly warns you that an alert won't stop anything.
It's the same logic as writing a sentence like this:
Y U gna be late
It's grammatically completely incorrect. But you can still understand it.
When it comes to chinese/japanese characters, many have the same phonetic reading. So you can do something similar, while selecting the wrong characters.
I think it's just easier for beginners (or teenagers) to go from phonetic to meaning. I guess advanced Chinese readers don't even read the words out loud in their head and go directly to meaning. I'm beginner/intermediate at Chinese and surprisingly, I noticed that my pinyin often seems better than many Chinese natives.
The real answer to your question: the most commonly used Chinese input method allows you to type the first pinyin letters only, and the algorithm will figure out the most likely Chinese characters you want.
It's not "the parents" can't read it. It's that people who don't use electronics have a harder time reading it.
Read the nonsense text aloud and then listen. Presumably with practice, you don't actually to actually speak aloud, and your 'inner monologue' voice is sufficient.
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.
> ends up costing 2-3% more than it would otherwise
This is a very simplified view. Cash handling is not free. Fraud levels with cash are different. Overall attractiveness of a small but cash-only business is different.
if the credit card merchant was able to extract 2-3% for using their network, what makes you think that a free debit card with low/zero cost wouldn't also be owned by the same credit card merchants and force the same fees to maintain their monopoly?
Europe has similar laws capping the fees on both credit cards and debit cards. We could do the same and it would work better for everyone except the credit card companies.
Because right now debit card purchases don’t have a transaction fee while credit cards do. I just had two recent purchases where the vendor charged a 3% surcharge for credit cards but would do the transaction by debit card free.
It is an example that greed is everywhere. Not only among card issuers but also among merchants. The moment merchant find a socially accepted way to extract extra they will do it. Similar to tip screens on square terminals in all takeout places. Costs nothing to ask, generates some additional revenue.
There is a reason every concert and sports venue near me has gone cashless. It's not because they enjoy giving away 2-3% of their revenues but rather those places aren't active every evening and tend to turn over employees quickly. Handling cash is expensive and risky in that environment.
In a concert and sports venue concessions environment you're also prioritizing throughput much more than most businesses. Being cashless helps with throughput a lot - the employees don't have to wait for people to count their money, then recount it, and spend time making change.
The linked to Atlantic article literally makes the opposite point; the subtitle there is "Blaming the housing crisis on hedge funds and private equity may be easy, but it’s dead wrong."
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.
