It's only local to China in the sense that they choose what {packages/CCP poison} go in there.

For all values of u, Linux is local to u in the sense that u chooses what packages go in there.

Ha ha, Linux already has..

# top -o gpu
top: unrecognized field name 'gpu'

.. d'oh! Never mind.

And is asking an AI to proofread something that much different than MS Word flagging spelling and grammar errors?

Great question, and I think there isn't any significant difference.

But LLMs use a type of math that apparently has the evil bit set. Meanwhile, up here in VirtueVille, the computer code we write to create things, uses different math that doesn't have the evil bit set. Conclusion: we can look down on those AI companies and their stupid users. ;-)

Now, I wouldn't expect an AI service to be better at copyediting than an average human, but I haven't kept up so I'm ignorant of their performance. (All my "AI" experience is that I used Claude twice to, it turned out, fix some typos in some RoR config initializers.) But whether an AI service is as good at copyediting as a human or not, it's that user's choice and, as long as they sign off on the work and really take responsibility for it (declare the words to be their opinion), then I've got no problem with them doing that, whether it uses LLMs or Markov chains or human slaves or Searle's Chinese Room.

The words will be judged by their quality, not provenance.

That said...

Here's a similar story where I'd probably come to the opposite conclusion(!), despite the clown in it pretty much doing what I preach above, though he was using the AI service to do a lot more than copyedit. Behold this 3M expert witness cross-examination (134 page PDF) in the Watson Grinding case. It's slow going at first, but eventually you'll be laughing and cringing like in a good episode of Curb Your Enthusiasm.

Fuck. I don't know.

Facial recognition is really neat technology, and we are all slightly better off thanks to its development. It's also imperfect, and nearly-guaranteed to be a net negative if you fully trust it, without human verification.

If you want to come out ahead by using cool technology, you're going to have to think about how to use it. There just isn't any shortcut around thinking.

If you have something which is only 99.9% reliable, that can be useful but it's going to be wrong sometimes. How are you preventing that wrongness from breaking your system? Insulting (much less turning-away) regular, recurring, paying customers is usually considered a really bad move, generally resulting in lower profits.

Free money falls out of the sky. Should we take it? Today we'll be interviewing Republicans, and learning that the answer is No because asceticism is a fundamental ideal that we should all strive toward. Burn a sacrifice instead!

As you increase the size of the bulk licenses' box, the amount of expensive cardboard increases as a square, but the number of licenses that fit inside the box increases as a cube. This results in a reduction of the amortized cardboard cost per license.

If a cost reduction isn't happening, then I would suggest people look into the shape of their bulk licenses. It may be that a same-sized box holds fewer licenses than it did ten years ago, due to the addition of "AI features" being tacked onto everything, resulting in irregular, difficult-to-efficiently-pack licenses.

It's classic Fed jawboning and double-speak. The problem is the Fed chairman can move market mountains with a few words.

If he had told everyone his plan was "interest rates are going up, AND we're expanding the Fed balance sheet", like what is happening now, people would have immediately started dumping bonds and tech stocks and we'd be way worse off now.

So he said the opposite, which just confused people and had zero effect. That's the goal...remain market neutral as much as possible. Which means they spew a lot of conflicting BS.

But this is a good thing. Meta patenting this means there's only one company we need to worry about pulling this shit.

Read his comment again. His objection has nothing to do with what's being patented; it's just a super-generic anti-recording complaint. This patent does not prohibit anyone else from pissing him off by recording things.

It doesn't help that the Facebook angle (that these are Meta's glasses rather than merely "smart glasses") injects all sorts of weirdness into the imagined application.

Facial recognition to identify people: has applications, such as the obvious "crap, I always forget names. Who is this guy, again?"

Highlight reels: WTF?! You call that an application? Extremely niche at best.

There's useful technology being discussed here, but apparently Meta doesn't want that to be the focus.

It's much worse than that; you would end up preventing anyone else from doing it for 20 years, until the patent expired. Of course that would make Slashdot!

Also, another consideration...input token cost.

I just made this mistake. I was in a long-running programming session with a lot of context, and I asked an unrelated question, "how to slob a knob" or something like that.

I constantly check my balance and noticed I burned 200k tokens with a simple question.

Then it dawned on me. It included every scrap of information in the session with the request, as it does, juicing me for a mountain of unnecessary input tokens.

So, lesson learned. For each unrelated task, start a new blank session. If you treat it like a never-ending text message contact, the context load (and input token costs) will snowball for no benefit.

The UI for Copilot kind of obscures this, but you click the "+" at the top and select "New Chat" for a fresh session.

Are you tweaking your reasoning level? That's becoming more important lately when managing your token burn.

I'm using Deepseek Pro until a better quality per cost option comes along, but I burned through about 100 million tokens per day ($8.70 at current prices) set at Maximum reasoning. You're basically telling it to torch tokens at ludicrous overkill, but the output is was exactly what I needed for the task. I finished a massive load of work in a few days that I had been stalled on for months because my other attempts at automating it were just not cutting it. And I'm too cheap and unemployed to pay for the equivalent horsepower at the usual frontier model rate.

Disclaimer: No, I'm not working on anything proprietary. Exposing proprietary code to a Chinese model may or may not be a good idea. Also they're about to double their prices.

Investors are chasing a company that achieves some kind of AI singularity. Let's set aside the fact that there's no reason to believe there is anything but diminishing marginal returns by making marginal refinements to current frontier models. Let's imagine someone hits the jackpot and gets, not even AGI, but a system that's as far ahead of today's frontier model are ahead of 2020's GPT 2.0 in performance.

Globally AI revenues are 150 billion, against a cumulative burn rate of 450 billion. A model that is a generation ahead of others would almost certainly capture the lion's share of that revenue.

If AGI magically appears as a Sam Altman has promised investors it will, a hundred million is way too low. Add, maybe, another zero to the revenues.

Conservatively, a safer assumption is that frontier models will get marginally better based on refinements in training and reinforcement and the other bits and bobs that go into these systems. The nobody is winning the lion's share of anything, at least overnight. But you have to define "safe". By "safe" I mean unlikely to lose money. But some investors are clearly defining "safe" as "having the greatest chance of owning a piece of the biggest thing ever."

I'm not following this super-closely, but if Anthropic is pursuing adding multi-step model based reasoning to their system, that could be the basis of a generational leap in capability. But if that is an approach that looks like it has a chance of working, then their competitors are no doubt pursuing the same thing. In that case you'd expect the revenue pie to grow as the scope of model utilty increases, but that growth to be split among several competitors. This could credibly result in a revenue stream for some of them that is as big as the entire industry's revenue stream today. But there's going to be hell to pay on the data center impacts end of things.

Is this something you do often?

No, it's pretty much one day a year (but usually several times on that day): spring cleanup (though my smizmar and I call it "truck day" (*)).

It's uncompostable/tainted yard waste, even furniture (if it's in truly horrible shape), and all sorts of other random trashy shit. When it happens, the amount of resulting trash is many times the size of our trash bin or what can even be put on on the curb for "large item pickup." The usual weekly process is overwhelmed and can't handle it; there simply isn't room. So we haul it.

* "Truck day" because I decided paying $72 to rent U-haul's big pickup truck for a few hours is preferable to paying $72k to buy a big pickup truck. And it usually gets used for more than just dumping trash; it's often a large-item shopping day too. (To be next decade's trash? Don't ask!)

Goddamn it! What's with this ticket about the CSV import being broken?!