4 points | by gpi 14 hours ago ago
4 comments
https://archive.is/5aegH
Oh cmon. I think calling this a puff piece is not accurate, this feels like fraud.
They run kimi3 on an “unsecured” sandbox, which had access to GitHub.
Kimi somehow knows the repo. It clones it. It finds the solution.
The hack? Using git?
https://blog.frontier.security/chinese-model-kimi-k3-breaks-...
First it was Anthropic, then OpenAI, and then Meta. Now the Chinese. I think the purpose ot these articles is to prove that "Hey! Look at how powerful our AI model is!" and gain a bunch of free publicity and buzz.
It's strange to announce capabilities with lab leaks.
https://archive.is/5aegH
Oh cmon. I think calling this a puff piece is not accurate, this feels like fraud.
They run kimi3 on an “unsecured” sandbox, which had access to GitHub.
Kimi somehow knows the repo. It clones it. It finds the solution.
The hack? Using git?
https://blog.frontier.security/chinese-model-kimi-k3-breaks-...
First it was Anthropic, then OpenAI, and then Meta. Now the Chinese. I think the purpose ot these articles is to prove that "Hey! Look at how powerful our AI model is!" and gain a bunch of free publicity and buzz.
It's strange to announce capabilities with lab leaks.