Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

K2 Horizon: A connected fleet of six open models

ifm.ai

Post reply on HN
Print view View original

141–144 of 144 posts

Re: K2 Horizon: A connected fleet of six open models

#141

Post by chme »

embedding-shape wrote: post #140
chme wrote: post #138

Earlier quoted context omitted.

The last point is pretty clear, and the text below clarifies it: > To protect the privacy of the individual and to strengthen the freedom of the information which concern the public the yet last point was added. The privacy of individuals is important, regardless where they store their private data. Their account information -- what they buy, their medical information and so on is stored on servers and could be hacke…

> The last point is pretty clear I think the second point is equally clear, and further up on the list. I agree that what people buy, their medical information and so on should be private, hence it should only be offline and not stored/handled on computers connected to the internet at all, the internet should be for public data exclusively, is my argument in the initial comment. Medical information would be only on e…

So you say it is okay when someone you trust, uses that trust to upload your personal files from your air-gapped system to the internet, for everyone to freely share it, because now the data is "public"?

To me private file stay private, even if they get leaked onto the open or closed internet, because the public has no right to know them, they are private data of an individual. They might no longer be secret, but they are still private.

Re: K2 Horizon: A connected fleet of six open models

#142

Post by embedding-shape »

chme wrote: post #141
embedding-shape wrote: post #140

Earlier quoted context omitted.

> The last point is pretty clear I think the second point is equally clear, and further up on the list. I agree that what people buy, their medical information and so on should be private, hence it should only be offline and not stored/handled on computers connected to the internet at all, the internet should be for public data exclusively, is my argument in the initial comment. Medical information would be only on e…

So you say it is okay when someone you trust, uses that trust to upload your personal files from your air-gapped system to the internet, for everyone to freely share it, because now the data is "public"? To me private file stay private, even if they get leaked onto the open or closed internet, because the public has no right to know them, they are private data of an individual. They might no longer be secret, but the…

I'm talking about liking waffles, then you appear asking why I hate pancakes. Fun.

> To me private file stay private, even if they get leaked onto the open or closed internet

We have very different definitions of what "private" means. Once it's leaked, it's no longer private, and pretending it can go back to being "private" after being on the public internet, is doing no one any favors.

Re: K2 Horizon: A connected fleet of six open models

#143

Post by chme »

embedding-shape wrote: post #142
chme wrote: post #141

Earlier quoted context omitted.

So you say it is okay when someone you trust, uses that trust to upload your personal files from your air-gapped system to the internet, for everyone to freely share it, because now the data is "public"? To me private file stay private, even if they get leaked onto the open or closed internet, because the public has no right to know them, they are private data of an individual. They might no longer be secret, but the…

I'm talking about liking waffles, then you appear asking why I hate pancakes. Fun. > To me private file stay private, even if they get leaked onto the open or closed internet We have very different definitions of what "private" means. Once it's leaked, it's no longer private, and pretending it can go back to being "private" after being on the public internet, is doing no one any favors.

I just find it very interesting that someone equates secrecy with privacy. In my opinion, and in the current law there is a distinction.

If private data got leaked, like revenge porn, it is a breach and that private data that belongs to an individual is still private, and still needs to be protected. This is what GDPR and other legislation is about. If secrecy and privacy is the same, someone that isn't able to protect their data sufficiently will not have any privacy, thus any leaking of data is now the fault of the person that got their data leaked, not of the person that broke the trust and leaked it.

Your conclusion seems rather extreme to me. So of course data that got leaked, and is no longer secret is still private, because 'private' means who should be in control of that data, not about if the person has control or not.

I also don't follow your point about waffles and pancakes, because this is a pretty big disagreement we have here. To me this dialog is more like you are saying "I don't like laws", and I say "While I agree that some laws are stupid, other laws are pretty useful, for instance people shouldn't be allowed rob other houses, even if they are able to do that or even where invited." And then, instead of agreeing, you sort of say, "No, I really mean that. If someone isn't able to defend their home properly or give out invitations to someone, it is okay to steal from them."

Re: K2 Horizon: A connected fleet of six open models

#144

Post by culi »

thepasch wrote: post #135
culi wrote: post #133

Earlier quoted context omitted.

This is absolutely not true. DeepSeek is most famous for publishing really in-depth papers on their training process but the other labs have started to do the same as well. If by "training corpus" you mean the actual data I already acknowledged that that's currently legally impossible. In fact everything I just said I said in my original comment. It's like you didn't read it at all. DeepSeek's GRPO Infrastructure, mu…

> If by "training corpus" you mean the actual data I already acknowledged that that's currently legally impossible. Did you read the site this very post links to? The entire point is that the training corpus, recipe, and scripts, as well as intermediate checkpoints, will be made available for K2 Horizon. Chinese models these days don't even release pre-trained weights anymore; all you get now is the finished post-tra…

Yes there have been several research projects like these that have fully open sourced their training data (mostly coming from Europe). But that's all they amount to. Experiments and projects.

The labs that are actually competing with frontier models are using data would usually be a violation of copyright to release openly.

> Chinese models these days don't even release pre-trained weights anymore; all you get now is the finished post-trained product.

No? That's absolutely not true. Qwen, GLM, Kimi, DeepSeek, etc all consistently release both the post-trained "Instruct/Chat" versions and the underlying "Base" (pre-trained) weights.

Which specific Chinese models are you thinking about?

Post reply on HN