onlinepersona
@onlinepersona@programming.dev
- Comment on Reddit’s deal with OpenAI will plug its posts into “ChatGPT and new products” 18 hours ago:
From what I understand LLMs are just large heuristic machines. They gather a lot of statistics on token order and return an answer to that with something that statistically should higher than other options. There’s no “understanding”. So to answer your question, no, they don’t understand the license.
Content is most likely scraped wholesale from websites, possibly run through some clean up to possibly filter out absolute garbage, and fed into an LLM to train it. An LLM can be tricked to reveal its training data (e.g repeat “fruit” forever). It’s in those cases where copyright infringement is detected and if action can and has be taken. There are court cases currently in review, the most popular being the one against Github Copilot for infringing on the license of sourcecode it ingested.
- Comment on Reddit’s deal with OpenAI will plug its posts into “ChatGPT and new products” 21 hours ago:
They might not even have to. I bet there are bots already having entire discussions by themselves on there.
- Comment on Google is redesigning its search engine — and it’s AI all the way down 1 day ago:
Yes please. Shoot yourself in the foot. It seems like it’s the only way people will even consider switching.
- Comment on YouTube Blocks Access to Protest Anthem in Hong Kong 3 days ago:
tbh, I know little about the capabilities of the Great Firewall. Maybe it already is possible to circumvent it with a VPN or an anonymity network like I2P or TOR. Also don’t know if they block per IP or in blocks. Possibly hosting the peertube instance on public cloud infra would make it difficult to block if the IP changed at certain intervals.
Hosting peertube could however provide dissenters with more options than youtube.
- Comment on YouTube Blocks Access to Protest Anthem in Hong Kong 3 days ago:
Use peertube?
- Comment on Fake WhatsApp and Instagram apps that can steal personal data 3 days ago:
So much to paying 30% to these stores for “security” and “app validation”.
- Comment on Kendrick be like 4 days ago:
Bunch of prions!
- Comment on Can somebody explain why game makers don't start their own companies together? 1 week ago:
I seriously hope so. Hopefully they make good games, become successful, and turn into the thing they hated.
- Submitted 1 week ago to gaming@beehaw.org | 59 comments
- Comment on Family matters 1 week ago:
Feel like that’s any expert job. Just get a roofer or electrician talking about their job and you’ll be lost too.
- Comment on What is the Anti Commercial-Al license and why do people keep adding it to their comments? 2 weeks ago:
Previous work has already shown that image generators can be forced to generate examples from their training data—including copyrighted works—and an early OpenAI LLM produced contact information belonging to a researcher
You don’t seem to be able to read the articles, yet have responded with junk anyway.
- Comment on What is the Anti Commercial-Al license and why do people keep adding it to their comments? 2 weeks ago:
Ironic, considering you are undoubtedly not a lawyer and have evidently never even dealt with copyright issues.
How is it ironic? I never said I was a lawyer, nor did I ever say I was giving legal advice, nor have I ever spoken with authority on the subject.
You however… can’t see the irony of your own statement.
- Comment on What is the Anti Commercial-Al license and why do people keep adding it to their comments? 2 weeks ago:
As a lawyer, what’s your opinion on the CC BY-NC-SA v4 license?
- Comment on What is the Anti Commercial-Al license and why do people keep adding it to their comments? 2 weeks ago:
Thanks for the support 💗
I’m not very affected by it, honestly. It makes me chuckle that people get so offended about a link in a comment. Sometimes I respond, but quite often I block and move on.
- Comment on What is the Anti Commercial-Al license and why do people keep adding it to their comments? 2 weeks ago:
I said so in my comment
So there is a big lawsuit against it
- Comment on What is the Anti Commercial-Al license and why do people keep adding it to their comments? 2 weeks ago:
How exactly do you expect to see the “source” of a language model?
- Comment on What is the Anti Commercial-Al license and why do people keep adding it to their comments? 2 weeks ago:
@Pacrat173@lemmy.ml the license is actually a Creative Commons license for Non-Commercial uses. Creative Commons is a copyleft license that’s “free to use with some restrictions”. Mostly used in art, literature, audio, and film, for my part I’m using it to license my comments. Anybody can cite with attribution, but commercial use is forbidden by the license.
The why: I just don’t like non-opensource commercial ventures. Google, Microsoft, Oracle, Facebook, Apple, and so on are harmful in many ways.
Enforcement and legality: Microsoft’s Github CoPilot (a large language model / “AI”) was trained on copyrighted text source code. A few licenses clearly state that derivatives should also be opensource, which CoPilot is not. So there is a big lawsuit against it. Many artists, non-programmer authors, musicians, and others are also unhappy that AI was trained on their copyrighted works and have sued for damages.
Until these cases make it out of court, it will not be clear if adding a license to comments could even jeopardize commercial AI vendors. - Comment on Men Use Fake Livestream Apps With AI Audiences to Hit on Women 2 weeks ago:
Wow… you are way too deep into whatever it is you’re into and are currently unable to see the shades of gray.
- Comment on Men Use Fake Livestream Apps With AI Audiences to Hit on Women 2 weeks ago:
Because people keep asking what the license is for.
- Comment on Men Use Fake Livestream Apps With AI Audiences to Hit on Women 2 weeks ago:
It’s because there are people who care about those things. There are people who are impressed by popularity, social status, etc.
- Comment on Men Use Fake Livestream Apps With AI Audiences to Hit on Women 2 weeks ago:
Well that’s definitely an… interpretation of what @BruceTwarzen@kbin.social wrote.
- Comment on neptune 2 weeks ago:
Probably a very good lens. If there’s no digital zoom and little digital correction, then quality can be quite good despite low pixel count.
- Comment on Twitter co-founder Biz Stone joins board of Mastodon's new US nonprofit | TechCrunch 2 weeks ago:
This is the enshittification fast-track, it feels like.
- Comment on The miracle of childbirth 2 weeks ago:
My only question is: why would they evolve like that?
Or alternatively, what the fuck were they before that gave birth like this?
- Comment on Explain yourselves, comp sci. 2 weeks ago:
- Comment on Explain yourselves, comp sci. 2 weeks ago:
It’s an array.
First time I heard of vectors in comp-sci was in C++. The naming still doesn’t make sense to me.
- Comment on Caption this. 2 weeks ago:
In order for AI to work, it needs a lot of training data. I personally have nothing against AI, just against the commercial variants whose models that aren’t made available to the public.
Of course just putting a license in text doesn’t provide automatic protection. It still needs detection of infringement and enforcement of the license. There’s an ongoing case against at least one commercial AI called Github CoPilot which could set a precedent for ignoring licenses.
- Comment on Caption this. 2 weeks ago:
Don’t put your dick in it…
- Comment on Caption this. 2 weeks ago:
Don’t mind me, just eating jelly with razor blades.
- Comment on What do you personally use AI for? 3 weeks ago:
Still, it does sometimes happen that they quote a source article verbatim.
I’m hoping it’ll quote the license I put in my comments (should my text ever be included in the training set) and gets somebody in trouble. But yeah, transformed anything is difficult undo to see what the source material was, so commercial LLMs can mostly just get away with it.