I personally don't count cached hits as $used... Neither in my harnesses, nor in the LLM-enabled apps I create. A cached token cannot be counted 1:1 as to a non-cached token, that would be silly.
Wait... when some Claude 5x/20x users say they are getting "$2000 of tokens for $100," does the 2k value include cached tokens, counted at the same $/token either way?
We cannot be this dumb as a community, can we? I must be wrong/misunderstanding..
I'm a fairly moderate user, never hit any kind of usage limits, but I used 44 million cache create tokens and 1.5 billion cache read tokens, which ccusage estimates would have cost $990, and calculates the different categories separately.
Don’t forget context. Basically I have 2 billion input and 1 million output. Every prompt you do, sends back the whole thing again and again. Let’s say you have 500k context used, you send 10 messages is 5 million. 100 messages 50 million.
Use 5 threats is 250 million.
But how is it even possible (bad harness?), or wise, to send 500k or 1M tokens per call? Regarding cache, how are you not hitting the 1hr cache? Also, start new chats early and often!
I have been "agentic coding" since Sonnet 3.5 and after this paper came out, it became my bible:
One one hand, it's political news, which I abhor on HN. On the other hand, international travel out of SF provides underlying support for the tech industry and without it, would cause deep headaches. On balance then, yeah, I don't mind this on HN.
From a nerd POV, this is what medium-range ballistic missiles look like when they fall on your city. These were designed for nuke MERVs, but were instead just used as Rods from God, as this is how desperate the RU war economy is currently. This is what is happening in Europe, right now.
Every time the imperialist Russians fire these, all nuke warnings go off everywhere, and we somehow avoid the nuclear apocalypse.
Super cool that this is now normalized and ignored.
But how do you judge the consequences? Either you have an infinite regress, or you end up declaring by fiat that some things are good and others are bad, just because. Which is...deontology--what "consequentialism" was supposed to be opposed to and an improvement on.
The concept that I think the world would be a better place if people of all political stripes could internalize is that nobody knows for sure what is good. We all have to make moral and ethical choices with incomplete knowledge. So when people make choices that you disagree with, the default presumption, at least, should not be that they're evil, but that they have different information than you do. That information can include different judgments about what is good and what is not. There is no single moral or ethical system that has all the right answers--including "consequentialism".
Neither Deontology, Virtue Ethics, nor Consequentialism describe the ends; only the tradeoffs.
You could have a deontological commitment to never giving a sucker an even break.
You could have a virtue ethicist who considers the Joker a paragon--I think some of them are in politics.
Consequentialism just says that deontology is too myopic, and locally following the correct rules is sometimes less good than maximizing long-term gains.
Consequentialism is ceteris paribus correct; but ceteris is often not paribus for humans, so pure consequentialism has a lot of footguns in it.
> Neither Deontology, Virtue Ethics, nor Consequentialism describe the ends
If you insist on just looking at the general, abstract terms as categories, instead of the actual ethical systems that are usually described as falling into those categories, I suppose that's true. But I don't see why it's relevant. In order to actually make ethical choices in the real world, you have to specify ends--your ethical choices have to bottom out at some point in saying that some things are good and some things are bad, just because. That's true whether you think you're doing Deontology, Virtue Ethics, Consequentialism, or what have you.
> Consequentialism just says that deontology is too myopic, and locally following the correct rules is sometimes less good than maximizing long-term gains.
And in making such claims, Consequentialism is both misdescribing Deontology and avoiding the actual issue.
First, there is nothing that restricts Deontology to "locally following correct rules". More generally, there is nothing that forbids Deontology from looking at consequences! Indeed, Deontology often requires you to look at consequences, since actions that might be innocuous taken in isolation can have serious ethical implications when put in context.
Second, when you say "maximizing long-term gains", what counts as "long-term" and what counts as "gains"? Any answer to such questions is going to bottom out, as I said, in claims that some things are good, and some things are bad, just because. There is no way to avoid that. But Consequentialism bills itself as avoiding that--as avoiding "just following rules" and looking at things rationally instead. And it doesn't and can't deliver on that promise. It just obfuscates what it's actually doing.
If you want to talk about ends, you're talking Axiology, not strictly Ethics.
By "ceteris paribus correct," I mean that if you were programming a superintelligent AI--and you knew exactly what you were doing, rather than structuring a learning schedule and feeding that a corpus--you would want a consequentialist.
Deontology and Virtue Ethics are patches for flaws in human morality. For example, the deontological rule "never kill the leader of the group and take over, even for the good of the group" is there because power is instrumentally useful enough that evolved social animals will deceive themselves about why they want power, so naive consequentialism doesn't work for them.
> If you want to talk about ends, you're talking Axiology, not strictly Ethics.
Even if I accept this quibbling over definitions, I don't see how it's relevant. We're talking about how to actually make choices. That's ethics.
> if you were programming a superintelligent AI--and you knew exactly what you were doing, rather than structuring a learning schedule and feeding that a corpus--you would want a consequentialist.
That if that you're glossing over is actually impossible, so I don't see how this is relevant either.
> Deontology and Virtue Ethics are patches for flaws in human morality.
No, they're recognitions of the fact that it's impossible for any finite being to compute and judge all of the consequences in real time, even if we assume there is some universally agreed on system for judging the consequences, which there isn't.
> power is instrumentally useful enough that evolved social animals will deceive themselves about why they want power, so naive consequentialism doesn't work for them.
Power is instrumentally useful for a superintelligent AI too, so "naive consequentialism" doesn't work for it either, at least not if you want us humans to survive in a world that has it.
When I was learning some homelab stuff, and was setting up pfSense, I was able to see the geos of all the scans/attacks on my home internet IP. I was surprised to see that Netherlands was up there with Russia and China in volume. They all got geo blocked.
What is it about the Netherlands that makes them so attractive to these people?
Amsterdam is one of the biggest (perhaps the biggest) global internet hubs - European equivalent of North Virginia - and also not a totalitarian country like Germany (otherwise there'd be more in Frankfurt).
I use Claude Code (Opus 4.6 at max effort) all day long, and I genuinely don't understand how this is possible. Is that usage paying off?
This is very likely due to my lack of understanding, but... how?
reply