so, (barely) reading between the lines of what Twitch's staff said:

1) they were forced by Amazon to add the AI training setting and to make it opt-out, because the overlords know perfectly well there's no reason anyone would ever opt-in

2) the actual Twitch staff who run Twitch do not like this and are of the same mind about it as all of you

3) Taking the time to opt-out, even if you don't really stream, gives them data to take to meetings with their overlords to prove that Twitch users dislike AI and pushing it too hard will drive people from the site.

Settings > Security and Privacy > Training for Generative AI

using zencastr for video podcast recording and it's excellent!

they just made a pile more stuff available on the free tier too, from 1080p to 4K

and shoved AI options in your face, because of course they did

i'm worried a bit because what's their corporate health like? that's a very expanded free tier. does anyone know anything?

for comparison, riverside charges $30/month for similar

This was my impression as well. Both people on this call are talking like there's somebody in a suit just off camera with one hand pointing at the sign that says "stick to the script" and the other white-knuckling the gun pointed at their careers.

https://mastodon.social/@glyph/117089201745379792

fucking impeccable emulation of a human being. is this a robot or a person? who could even say. truly you can't tell if they're using an LLM, no point in trying to detect it. chefs kiss. no notes

I can't _know_ what is going on in people's hearts, but my read is somewhat more charitable to the people in the video here. It's wild to me that they are people with this much authority who seem helpless, but rather than "callous jerks who don't care what you think" their facial expressions and body language read to *me* as people who have been ordered to implement a harmful idea and are struggling mightily to do it in the least harmful way they can https://toot.majorshouse.com/@majorlinux/117085014811292459

Another reason I try my darndest to avoid LLM-built "FOSS" tools: the humans who publish them often enough aren't prepared to support them.

When that neat utility that Claude burped out is deprecated by some breaking change, are you going to have someone to reach out to? Are you going to understand it? Or will the "author" have moved on toward some other thing that nabbed their attention?

okay DNS fanciness engaged, finally https://bsky.app/profile/glyph.im

@whbboyd @glyph Hilarious.

If we were to assume you were typing actual content, and the standard paperback page contains around 1500 characters and books are 300 pages… you’re writing 377 books per second.

It would also be ~1.3 million kilograms of TNT equivalent per second from your calculations.

Which, using the most ridiculous possible unit, is about 3500 kilograms per book-second of TNT.

@glyph Okay, I can't resist.

So the overhead for the HID protocol is 8x for a single keypress. (I don't think the HID protocol can actually saturate a 10Gb/s link, but we're going to ignore that for now.) That means for a stream of single keystrokes, you're sending just under 170 million per second. (You can send more data by chording, up to six bytes per poll, but you won't get anything out of pressing individual keys faster than that.) We'll optimistically assume that your keyboard has 102 keys (as we know it doesn't) and that you're pressing each of those 102 keys equally often (as we know you're not). That means each key is getting pressed about 1.6 million times per second.

By the standardized definition of WPM for typing, this is 329 thousand words per minute.

I dunno exactly what the throw of your keyswitches is, but if it's a standard-ish 4mm, then in order to be pressed in 300 nanoseconds, it will need an average speed of 47 thousand km/hr. Under constant acceleration from your fingertip, it will accelerate at 86.5 billion m/s², bottoming out at 94.6 thousand km/hr and delivering some 350 kJ to the backplate. (This is about the energy in 75 grams of TNT.) The force required for this keypress is 86 meganewtons, or 8.8 metric tonnes. This *may* result in breaking a fingernail, or possibly some finger strain. (Also, the return spring will have to be as strong, meaning the press force will be even greater.)

For those of you that are fucking off to Codeberg forever and making forks of the Linux kernel and everything else, godspeed, I hope that you are successful and I hope your efforts send a useful signal. I hope that I have the energy to help out a little. But for those of you who are watching everyone in your community go a *little* bit nuts, I don't think you need to despair that we are descending into a permanently unfixable mess. The smaller fights are also worthwhile.

Many projects that stop will not need to do a giant dramatic remediation; they will have been paying attention all along, they will have been *trying* to get good utility out of the LLMs, they will not have their critical faculties _instantly_ obliterated (it is indeed the creeping and subtle nature of the problems which makes them so problematic).

A lot of "slop" is just trivial fixes that cannot possibly be copyright-infringing or holding hidden security holes. Vigilance in human review does tend to decrement over time, but that's not the same as it being totally worthless. Projects will still have standards, even if they're a little degraded. Some projects will simply observe these things through direct experience and eventually stop.

This doesn't mean that there are no boundaries. I will strenuously object if you smoke around me; if you're jonesing, and we're outside, I might begrudgingly bear it for the benefit of conversation. I will absolutely forbid you to smoke around my kid, though, and if you try to, we might not be friends for much longer. This distinction highlights that while all smoking is bad, not all smoking is *the same*.

It's not like the consequences aren't serious; some people will, in fact, contract lung cancer. Some of them will die from it, often in hideous agony. So I don't want to downplay that "it's fine to smoke" or anything.

But neither is a smoker who occasionally condemns others to secondhand smoke a morally irredeemable person, nor are they a doomed lost cause. They could always stop. It's easier to not-start than it is to stop, but lots of people stop.

I have a very nasty reaction to tobacco smoke. I might be allergic to it, but in any case, inhaling even a tiny bit is a DEEPLY unpleasant experience for me. Being around it nauseates me. And obviously it's not healthy for the person smoking it. Indeed, it's usually, eventually, lethal.

And yet, I have many friends who are, or have been, smokers. Most of them smoked socially. A few were chain smokers. Some did in fact blow smoke in my face on occasion. I survived. So did they.

After a couple of epic crash-outs the last few days, today I am feeling a bit more hopeful, which I assume is going to get more people to yell at me:

Your favorite open source project that is "embracing slop" is not lost forever. Some will be, but I think overall the ecosystem will be eminently recoverable.

If I may, here's another metaphor. 🧵

Apparently delivery drivers are getting $31 an hour minimum rate guarantee, starting on Monday. Seems like a lot but it does not cover waiting time or costs (fuel and maintenance) but better than my current rate of $20ish gross per hour. Anyway. I am glad it is happening and if you think this is going to put the cost of deliveries up, yes it will and GOOD. You were getting food delivered by underpaid labour before. Thank you #union #deliverator

Co-worker: I completed the project ahead of schedule with this simple hack

Me: I told you, I’d rather you didn’t refer to me that way

@glyph this has been a life-saver. It does go through batteries faster than I'd like, but it's super fast and hasn't misread anything.

https://treedix.com/products/treedix-usb-cable-tester-usb-c-cable-tester-usb-tester-for-data-transmission-and-power-transmission-cable-testing-resistor-testing-type-c-emarker-for-usb-a-micro-b-micro-b-3-0-type-c-lighting-mini-b