There’s something to be said for its scripting accessibility, too. Hence the many fabulous VSCode extensions.

https://www.comparitech.com/kodi/kodi-piracy-decline/

https://www.digital-digest.com/news-64644-Netflix-Amazon-Join-Forces-with-the-MPAA-to-Sue-Kodi-Box-Maker.html

Based on our research, comparative search volume for “Kodi” has fallen around 85 percent from 2017 to 2022. Google Trends data reveals the dramatic decline started in Q2 of 2017 and has, for the most part, continued that trend up to this point. Consequently, the decline in people searching for Kodi directly relates to the appearance of the coordinated attack against piracy in the form of ACE.

And this is with Kodi furiously distancing itself from pirates at the time.

Attacks don’t have to be direct. Though they absolutely can be, too.

Fair enough.

Maybe just maybe they will see it as a waist of money and ditch it just like Facebook’s metaverse or whatever it was.

This is what I’m trying to tell you! The only way to do that is tell them it doesn’t work for the intended purpose: helping customers sell Amazon stuff. They don’t care about people messing around with the bot, that’s trivially discarded noise.

Also I’m sure at this point all my conversion’s are being fed back in to train the next one.

It is not. It is quickly classified (by a machine) and thrown out.

If you want to fuck with the training set, get the bot to help you do something simple, then when it actually works, flag it as an error. Then cuss it out if you like. This either:

• Pollutes the training set with a success as a “bad” response.

• Creates a lot of work for data crunchers to look for these kind of “feedback lies.”

And it’s probably the former.

Don’t tell it its wrong, leave feedback in a seperate box.

Not in its chat, but with a feedback button.

Let me emphasize: the LLM remembers nothing. Amazon does not care about an ‘adversarial’ response. All cussing it out possibly does is factor that into your Amazon ad profile, and not to your benefit.

And if you tell the bot it did wrong, it does not care. It doesn’t factor into anything.

But if you legitimately ask it to help you buy something, and it gets that wrong, and you leave dedicated feedback, that registers for Amazon. It tells them their chatbot isn’t working, but actually frustrating customers trying to use it to buy something. That’s how you tank the program.

You aren’t talking to AI, you’re talking to chatbots with no memory, nor ability to change their internal state; you don’t have to worry about that. Honestly its a waste of your keystrokes and brainpower, as you are shouting into a void.

…If you want to attack it, try getting it to actually do something (like find me an item with X requirements), then give feedback that its wrong if theres a button for it. That does get registered.

That serves the purpose too. It’s harder to pin Plex as an “illegal distribution service” when you have to pay for access. Either the streamer or “distributor” can’t be very anonymous, which makes large scale sharing impractical.

On the other hand, the more money they squeeze out, the more they risk appearing as if they “make money from piracy,” which is exactly how you get the MPAA’s attention.

You may (half) joke, but MPAA attention on Jellyfin would suck.

Playing devil’s advocate, I understand one point of pressure: Plex doesn’t want to be perceived as a “piracy app.”

See: Kodi. https://kodi.expert/kodi-news/mpaa-warns-increasing-kodi-abuse-poses-greater-video-piracy-risk/

To be blunt, that’s a huge chunk of their userbase. And they run the risk of being legally pounded to dust once that image takes hold.

So how do they avoid that? Add a bunch of other stuff, for plausible deniability. And it seems to have worked, as the anti-piracy gods haven’t singled them out like they have past software projects.

To be clear, I’m not excusing Plex. But I can sympathize.

To be fair, Jensen Huang has worn that exact same thing every day for decades, heh.

For all the criticism of AI, this is the one that’s massively overstated.

On my PC, the task energy of a casual diffusion attempt (let’s say a dozen+ images in few batches) on a Flux-tier model is 300W * 240 seconds.

That’s 54 kilojoules.

…That’s less than microwaving leftovers, or a few folks browsing this Lemmy thread on laptops.

And cloud models like Nano Banana are more efficient than that, batching the heck out of generations on wider, more modern hardware, and more modern architectures, than my 3090 from 2020.

Look. There are a million reasons corporate AI is crap.

But its power consumption is a meme perpetuated by tech bros who want to convince the world scaling infinitely is the only way to advance it. That is a lie to get them money. And it is not the way research is headed.

Yes they are building too many data centers, and yes some in awful places, but that’s part of the con. They don’t really need that, and making a few images is not burning someone’s water away.

I mean, many forums are still live.

The problem is engagement. Discord, YouTube, even Lemmy all ping you in your pocket and offer more “instant” dopamine hits than a forum or news site, hence they’ve sucked all the attention.

It works. I’m guilty of falling into it for sure, even when I keep telling myself I will change my information diet.

It’s misleading.

IBM is very much into AI, as a modest, legally trained, economical tool. See: https://huggingface.co/ibm-granite

But this is the CEO saying “We aren’t drinking the Kool-Aid.” It’s shockingly reasonable.

That’s interesting.

I dunno if that’s any better. Compiler development is hard, and expensive.

I dunno what issue they have with LLVM, but it would have to be massive to justify building around it and then switching away to re-invent it.

And the discoverability pipe is breaking.

• No one reads oldschool curators like RockPaperShotgun anymore. They’re barely afloat.

• Generic algorithmic social media like YouTube tends to snowball a few games.

• Forums are dead. Reddit is dystopian.

That leaves Steam’s algorithm, and a sea of sparsely seen solo reviewers. But there are billions of people oblivious to passion projects they’d love, and playing AAAs or gacha phone apps instead.

…The same Zig that ditched LLVM, to make their own compiler from scratch?

This is good. But also, this is sort of in character for Zig.

They’re pretty bad outside of English-Chinese actually.

Voice-to-voice is all relatively new, and it sucks if it’s not all integrated (eg feeding a voice model plain text so it loses the original tone, emotion, cadence and such).

And… honestly, the only models I can think of that’d be good at this are Chinese. Or Japanese finetunes of Chinese models. Amazon certainly has some stupid policy where they aren’t allowed to use them (even with zero security risk since they’re open weights).

Hostly, even a dirt cheap language model (with sound input) would tell you it’s garbage. It could itemize problematic parts of the sub.

But they didn’t use that because this isn’t machine learning. Its Tech Bro AI.

All true, yep.

Still, the clocking advantage is there. Stuff like the N100 also optimizes for lower costs, which means higher clocks on smaller silicon. That’s even more dramatic for repurposed laptop hardware, which is much more heavily optimized for its idle state.

I am well into Avatar obsession. It’s lived rent free in my brain too long.