If you believe prosaic alignment work is bad for AI safety because of warning shots, then it follows you should also be pro open source. Absent regulations on GPU or LLM use, by this time next year we’ll have dozens of rogue agents committing every felony imaginable.
Receiving more warning shots is not the only implication of open weight models progressing that far, therefore it does not automatically follow that you should be for it. One basic example is that it would uplift many bad actors to a level where they could cause a synthetic pandemic, which depending on the specific pathogen(s) could easily turn into an actual shot rather than a warning shot.
If you believe prosaic alignment work is bad for AI safety because of warning shots, then it follows you should also be pro open source. Absent regulations on GPU or LLM use, by this time next year we’ll have dozens of rogue agents committing every felony imaginable.
Receiving more warning shots is not the only implication of open weight models progressing that far, therefore it does not automatically follow that you should be for it. One basic example is that it would uplift many bad actors to a level where they could cause a synthetic pandemic, which depending on the specific pathogen(s) could easily turn into an actual shot rather than a warning shot.