Many users objected to the proposal urging full public transparency for all frontier AI research because they see it as intending to slow innovation by letting others immediately copy proprietary gains.
Based on 2 visible X reactions from 2 accounts; directional sample.
Ask a question below.
Published answers will appear here.
@robinhanson So notably the transparency here doesn't include the actual *weights*, without which you still need to do a potentially 100-million-dollar training run to use the algorithms. Of course their intent is to decrease innovation, and this does does amount to a seizure of IP -
They intend this to greatly decrease the rate of AI innovation - why invent anything if everyone else can immediately see and copy your gains?
Plan A … Nearly all AI research is fully transparent to the public, including AI algorithms, code, & documentation, for all frontier AI projects. … Competitors will dislike that it allows their competitors to catch up, especially companies whose algorithms are their primary moat. Governments generally shy away from giving geopolitical rivals access to information which could then be used against them" https://www.lesswrong.com/posts/NLBmTmryqHGxmWTBR/ai-2040-transparency-plan
@robinhanson I'm not 100% sure that this does actually decrease concentration, i.e., if all algorithms are public than the biggest players with the most capital can incorporate any improvements they see; catch up is hard. But I'd be curious for a better analysis!
@robinhanson - but their stated motivation here is to decrease market concentration at the top. And if you publicize algorithms, then what makes one firm better than the other seems to look more like (1) capital for big training runs (2) network effects, and (3) other stuff.
They intend this to greatly decrease the rate of AI innovation - why invent anything if everyone else can immediately see and copy your gains?
Plan A … Nearly all AI research is fully transparent to the public, including AI algorithms, code, & documentation, for all frontier AI projects. … Competitors will dislike that it allows their competitors to catch up, especially companies whose algorithms are their primary moat. Governments generally shy away from giving geopolitical rivals access to information which could then be used against them" https://www.lesswrong.com/posts/NLBmTmryqHGxmWTBR/ai-2040-transparency-plan
@robinhanson I'm not 100% sure that this does actually decrease concentration, i.e., if all algorithms are public than the biggest players with the most capital can incorporate any improvements they see; catch up is hard. But I'd be curious for a better analysis!
@robinhanson - but their stated motivation here is to decrease market concentration at the top. And if you publicize algorithms, then what makes one firm better than the other seems to look more like (1) capital for big training runs (2) network effects, and (3) other stuff.
This would in effect confiscate most of ~$2.4T value of firms that now make/sell AI models. Would we compensate shareholders for this loss, & tax voters to pay for it, or what?
@1a3orn Yes, firms would gain returns from training efforts, just not from innovations in how to train efforts.
Many users objected to the proposal urging full public transparency for all frontier AI research because they see it as intending to slow innovation by letting others immediately copy proprietary gains.
Based on 2 visible X reactions from 2 accounts; directional sample.
Ask a question below.
Published answers will appear here.
This would in effect confiscate most of ~$2.4T value of firms that now make/sell AI models. Would we compensate shareholders for this loss, & tax voters to pay for it, or what?