Re: Concerns about AI code in the kernel

[Ellie]

On 5/13/26 10:17 AM, Ellie wrote:
> It also seems like if an LLM was made with properly licensed data

I just realized I should clarify what such data might look like:

I'm not a lawyer, but I'm guessing from a moral practical angle the 
expectation would be that the training data for an LLM would have a 
license that 1. doesn't require attribution and 2. is considered 
compatible with relicensing to the kernel's GPL.

(Whether this approach would safely avoid legal problems I wouldn't 
know. But it seems to me like it'd cause less upset in the community.)

The majority of public projects on Github, apparently those are what 
Co-Pilot is trained on, seem not to meet above requirements. I'm 
guessing for Claude Code the situation might be similarly concerning.

Regards,

Ellie