Hacker Newsnew | past | comments | ask | show | jobs | submit | mixtureoftakes's commentslogin

There is zero chance models are not trained on pirated media and games, especially most popular ones.

Makes complete sense to study things people actually want to use and consume


I agree 100% with sentiment. It is inconceivable to me that models were not trained, either directly, or indirectly through other aggregates, on code and binaries and, of course, media, fetched from torrent or other filesharing sites.

At least for Google, they have almost every game worth playing documented in precise frame by frame detail on YouTube. They don't need access to the source code, or compiled binaries to reverse engineer game mechanics. They built vision models perfectly capable of ingesting those game play videos and derive game play understanding from them.

Plus, there's so much open source, and game engine SDKs, freely available, that with a few game engineers at Deep Mind, they can write training systems that connect the dots.

It's definitely a gray area because these models are doing what an indie game dev would do, but at a scale and precision that's basically pirating the game, and getting around the rules of copyright to do it legally.


wouldn't that information leak out or somehow found out? too many people already involved. With some, like Seedance it's obvious of course.

I thought the fact they fed copyrighted material into the models has been established.

It has leaked out as part of court cases that Meta has trained LLMs using pirate text torrents.

No reason to belive none of the labs have similarly trained on binaries.


Meta dropped out of the LLM race for a while shortly after the court rulings came out. (and their LLM sucked relatively to other offerings) Then Zuckerberg installed a new AI guy who specialized in things like data labelling. Then Meta came back with a better LLM.

Can't say for sure, but the coincidences align pretty well.


It will once all those involved cash out from the AI companies they’re working with…

Also known as a singular 200$ subscription. Nobody sane is paying "real" api prices

Companies do and for grunt sw jobs even using opus is relatively cheap.

Which companies do that? It is way cheaper to get enterprise team plan.

I know at least about one. I suspect that though that it will be cheaper at the beginning. Now it's almost 100$k per month for a company, which I would be surprised if hit with subscriptions.

It's really hard to read it's just unusual which requires focus, and in turn is much more enjoyable

Agreed.

Most whiplash i've experienced from a personal website this year. Good job

Hey just curious, to what extent you using ai to help with proactive exploit discovery?


We use frontier AI models to review all our code but it hasn't replaced human review and humans need to review the output from it. It adds significantly more work and has resulted in a lot of issues being fixed.

We're not currently putting a large amount of resources into finding Linux, AOSP or Chromium vulnerabilities this way because many others are doing it with far more resources including Google. There are much higher impact things for us to focus on instead. Our focus remains systemic privacy and security improvement rather than case-by-case fixing of bugs. We definitely do that much better through shipping Linux and Android patches much faster than the Pixel OS along with fixing many issues ourselves but it's not our priority for the most part. Certain parts of it are priorities such as finding and fixing VPN leaks because Google doesn't consider it to be a relevant security issue.


>Coding flattened out like 9 months ago.

It absolutely has not. You do have a nice point with data though, somehow haven't thought about that


not taking risks is the biggest risk in the times when the most anticipated game drop (gta 6) is probably going to be decompiled within days if not hours of release


I wish there was a way to sort by recency, if i'm mainly interested in finding obscure gems from like last 6 to 12 months of ai and tech updates


added a few minutes ago!


Why's the newest video 5 months old?


Even with a fresh dump, it would lag behind YouTube, since a video only gets in once someone links it on HN.

I took all the channels HN keeps recommending (3+ videos in Karpathy's data) and made a page with their fresh uploads as they come in: https://hn.youtldr.com

(I make YouTLDR, which writes the one-line answer under each video.)


Don't see it.


first do a search, then on top you'll see "Filter". might try to make more prominent from main page


Yes, please, I couldn't figure it out as well.


added! now can be done via Filter on the main page.


thanks!


Yeah, this is the first thing I looked for and when I couldn’t find it I easily decided it’s something I would never come back to. Add this feature and it’s something I would come to frequently


plus one when in programming mode; i have playlists on YT; but this helps to focus... and mostly I find HN has very interesting things that spawn a ton of thoughts...

people like us don't have that much time...


Seconded! (please)


Well, build your own in 2.1hrs (or 2 minutes if you have access to the source)


Yeah i'm pretty sure this is doable multiple times on a 20 dollar subscription, just wow


Not worth (I'm an insider)


your username's not orange.


Less competition? :D


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: