Mama told me not to come.
She said, that ain’t the way to have fun.
And this is why I hate those laws that are intended to protect kids. Yeah, it would be nice if kids couldn’t see stuff they shouldn’t, but it’s even better if my PII isn’t stolen. I’d rather my kids accidentally see porn once in a while than for their identity to be stolen.
By quality I meant resolution, I don’t need 4k, but I do need specific shows my wife and kids like.
I have a NAS set up with some movies and whatnot, so I’ve talked to my wife about setting up a budget to purchase content we want and then cancelling our streaming services. So we’d be limited to what’s available on DVD/Blu-Ray, but most of what my wife and kids watch are still available there.
The cost isn’t the issue, I really hate ads and I’m worried ad-free tiers will go away (or become unreasonably expensive).
I disagree that it needs to be explicit. The current law is the fair use doctrine, which generally has more to do with the intended use than specific amounts of the text/media. The point is that humans should know where that limit is and when they’ve crossed it, with motive being a huge part of it.
I think machines and algorithms should have to abide by a much narrower understanding of “fair use” because they don’t have motive or the ability to Intuit when they’ve crossed the line. So scraping copyrighted works to produce an LLM should probably generally be illegal, imo.
That said, our current copyright system is busted and desperately needs reform. We should be limiting copyright to 14 years (as in the original copyright act of 1790), with an option to explicitly extend for another 14 years. That way LLMs can scrape comment published >28 years ago with no concerns, and most content produced >14 years (esp. forums and social media where copyright extension is incredibly unlikely). That would be reasonable IMO and sidestep most of the issues people have with LLMs.
Let me know if it works and I’ll follow. I don’t need quality, I just need something for my kids to watch occasionally.
Consider PiHole as a whole home network first line of defense.
And that’s exactly what that page discusses. It links three options you can try:
The first two are paid, the last is FOSS, and it claims each can mount Backblaze B2 as a Windows drive. I haven’t tried any of them, so YMMV.
That depends on how similar your resulting algorithm is to the sources you were “inspired” by. You’re probably fine if you’re not copying verbatim and your code just ends up looking similar because that’s how solutions are generally structured, but there absolutely are limits there.
If you’re trying to rewrite something into another license, you’ll need to be a lot more careful.
I complain all the time. But that’s not the subject of this post…
IP Man. Great movies.
I’m not going to be monitoring Chinese code projects. They don’t seem to care much about copyright, so they’ll probably just yoink the code into proprietary projects and not care about the licenses.
What am I going to do, sue someone in China? And decompile everything that comes from China to check if my code was likely in it? That’s ridiculous. If it’s domestic, I probably have a chance, but not if it’s in another country, and especially not one like China that doesn’t seem to care about copyright.
Here are options for to mount Backblaze B2 as a drive. It’s $6/TB/month, and I think they allow <1TB, so for 300GB you’d pay ~$2/month. So I think they’re pretty competitive, but I’m not familiar with Google Drive’s terms. They’re certainly in the same ballpark, if not cheaper, but it depends on your egress and Google Drive’s policies around that (how much you download from their service).
I’m pretty sure I do understand the issue. Here are some facts (and an article to back it up):
And here’s my interpretation/guesses:
So:
Using socketed RAM won’t fix performance issues related to running out of RAM, that issue is the same regardless. Only adding RAM will fix those performance issues, and Apple could just as easily make “special” RAM so you can’t buy socketed RAM on the regular market anyway (e.g. they’d need a different memory standard anyway due to Unified Memory).
I have hated Apple’s memory pricing for decades now, it has always been way more expensive to add RAM to an Apple device at order time vs PC competitors (I still add my own RAM to laptops, but it’s usually way cheaper through HP, Lenovo, etc than Apple at build-time). I’m not defending them here, I’m merely saying that the decision to use soldered RAM makes a lot of engineering sense, especially with the new Unified Memory architecture they’re using in the M-series devices.
The built-in Digital Wellbeing & Parental controls works. I have it on my Android 11 device, haven’t tested on anything newer (it’s not on my Graphene OS device based on the most recent Android though).
Settings > Digital Wellbeing & Parental controls > Dashboard > click the timer icon next to an app and set a limit
If you want something outside of the Google ecosystem (e.g. you’re running GrapheneOS), the following should work (untested):
There are probably others, that was just a cursory check.
And how will you know what original work(s) to compare it to?
code generated by an AI is arguably not a “substantial portion” of the software
How do you verify that though?
And does the model need to include all of the licenses? Surely the “all copies or substantial portions” would apply to LLMs, since they literally include the source in the model as a derivative work. That’s fine if it’s for personal use (fair use laws apply), but if you’re going to distribute it (e.g. as a centralized LLM), then you need to be very careful about how licenses are used, applied, and distributed.
So I absolutely do believe that building a broadly used model is a violation of copyright, and that’s true whether it’s under an open source license or not.
It certainly can. Most licences require derivative works to be under the same or similar licence, and an AI based on FOSS would likely not respect those terms. It’s the same issue as AI training on music, images, and text, it’s a likely violation of copyright and thus a violation of open source licensing terms.
Training on it is probably fine, but generating code from the model is likely a whole host of licence violations.
This source seems to indicate that’s not the case:
Google Search & Other (56.93%)
2023 Total Google Search & Other Revenue: $175.04 billion
This is revenue generated primarily from ads shown on Google’s search results pages and other search-related services.YouTube Ads (10.26%)
2023 Total Youtube Ads Revenue: $31.51 billion
This is revenue from ads shown on YouTube videos, including display ads, overlay ads, skippable video ads, and non-skippable video ads.Google Network (10.20%)
2023 Total Google Network Revenue: $31.316 billion
This is revenue from ads displayed on websites and apps that are part of Google’s ad network, beyond Google-owned properties.Google Other (11.26%)
2023 Total Google Other Revenue: $34.68 billion
This is revenue from Google’s other ventures and products, such as hardware (like Pixel phones and Nest devices), Play Store purchases, and other non-advertising sources.Google Cloud (10.75%)
2023 Total Google Cloud Revenue: $33.08 billion
This is revenue from Google’s cloud computing services, such as computing power, storage, and data analytics offered to businesses and developers.
So, 57% from search, and only 10% from ads on non-Google pages.
Looks like I’ll need to set up pihole then. Thanks for the info!
I’m guessing furries.
Sure, but that’s not what LLMs are doing. They’re breaking down works to reproduce portions of it in answers. Learning is about concepts, LLMs don’t understand concepts, they just compare inputs with training data to provide synthesized answers.
The process a human goes through is distinctly different from the process current AI goes through. The process an AI goes through is closer to a journalist copy-pasting quotations into their article, which falls under fair use. The difference is that AI will synthesize quotations from multiple (many) sources, whereas a journalist will generally just do one at a time, but it’s still the same process.