The blog post critiques the response of the free and open-source software (F/OSS) community towards the use of their code for training proprietary language models, arguing against a strategy of withdrawal and denial. The author agrees that AI companies disrespect F/OSS developers by using their work without consent but believes the solution lies in reclaiming the technology rather than rejecting it. They propose that instead of blocking access to code, developers should push for models trained on their work to be free and accessible, countering the trend of privatizing collective knowledge. The discussion shifts to the reality of LLMs changing the programming landscape. Salvatore Sanfilippo, known as antirez, suggests adapting to these changes rather than resisting them, viewing LLMs as democratizing technology. While the article acknowledges this perspective, it questions who ultimately benefits from the models produced. It highlights the importance of keeping the collective labor of F/OSS developers as a shared resource rather than allowing it to become proprietary. In terms of licensing evolution, the article draws parallels between past F/OSS licensing challenges and today's situation with AI training. Historical licenses like GPLv2 and GPLv3 emerged in response to exploitation by companies, adapting to technological changes. The author envisions a new licensing framework, potentially a "Training GPL," that explicitly permits the use of F/OSS code for AI training while mandating that models derived from this training remain open and accessible. This reflects a materialist approach, suggesting that the community must evolve its tools to match the new realities of AI technology.