What I'm struggling with on this case of open source is that it is very hard, as I understand it, to critique the network generated by LLMs. Meaning, with most opensource software you can review the code and identify vulnerabilities. But because LLMs are for the most part generated through data acquisition and refinement, it is hard for an opensource contributor to review the data and identify inconsistencies, threats. Have you heard thoughts on how that can be addressed?
That’s a great question and one that cuts through all the open source optimism we’ve been seeing since DeepSeek emerged. I wonder about tracking tokens in LLMs from a copyright standpoint too, for the original creators. Maybe Dre Labre @designfiction has some thoughts on this. We’ve been talking about open source and data authentication lately.
What I'm struggling with on this case of open source is that it is very hard, as I understand it, to critique the network generated by LLMs. Meaning, with most opensource software you can review the code and identify vulnerabilities. But because LLMs are for the most part generated through data acquisition and refinement, it is hard for an opensource contributor to review the data and identify inconsistencies, threats. Have you heard thoughts on how that can be addressed?
That’s a great question and one that cuts through all the open source optimism we’ve been seeing since DeepSeek emerged. I wonder about tracking tokens in LLMs from a copyright standpoint too, for the original creators. Maybe Dre Labre @designfiction has some thoughts on this. We’ve been talking about open source and data authentication lately.