4 ms·
> copyright and license laundering The nice thing is that this works in both directions. Just as Microsoft and Co. can train on people's data, nothing stops th
by grumbel 4y ago
> copyright and license laundering
The nice thing is that this works in both directions. Just as Microsoft and Co. can train on people's data, nothing stops the people from training on Microsoft's chatbot output.
As long as we don't get any lawsuits derailing this practice, we actually might end up fine, as there is nothing that'll allow anybody to get ahead in AI without automatically provide a mountain of training data for the competition.
If that means the end of copyright right as we know it, I am all for it.
- marginalia_nu 4y agoMicrosoft &co doesn't (to my knowledge) train their LLMs on their internal source code.
- tough 4y agoWindows has several codebases leaked no? Let's train a windows XP source code model for teh lulz
- gpderetta 4y agoinbefore corps push for AI personhood to make them sign NDAs
- marginalia_nu 4y agoI honestly don't think Microsoft puts much that value in source code that's from around the time Brendan Fraser was doing Mummy movies.