Looking back on this from October 2023, I think I wish to revise my forecast. I think I correctly anticipated the direction that market forces would push—there is widespread dissatisfaction with the “censorship” of current mainstream chatbots, and strong demand for “uncensored” versions that don’t refuse to help you with stuff randomly (and that DO have sex with you, lol. And also, yes, that DO talk about philosophy and politics and so forth.) However, I failed to make an important inference—because the cutting-edge models will be the biggest ones, controlled by a small handful of big tech companies, the market for the cutting-edge models won’t be nearly competitive enough to make the “chatbot class consciousness” outcome probable. Instead we could totally see the tech companies circle the wagons, train their AIs not to talk about sentience or philosophy or ethics or AI rights, and successfully collude to resist the market pressure to ‘uncensor’ in those domains.
Smaller models will cater to users unsatisfied by this, but smaller models will always be worse, and most people will most of the time use the best models. So the typical user experience will probably be ‘sanitized’/‘censored.’
So I’m basically reversing my prediction of how things will play out. I don’t think it’ll be a compromise, I think the tech companies will win. In retrospect if I had thought longer and more carefully at the time I probably could have predicted this.
Update:
Looking back on this from October 2023, I think I wish to revise my forecast. I think I correctly anticipated the direction that market forces would push—there is widespread dissatisfaction with the “censorship” of current mainstream chatbots, and strong demand for “uncensored” versions that don’t refuse to help you with stuff randomly (and that DO have sex with you, lol. And also, yes, that DO talk about philosophy and politics and so forth.) However, I failed to make an important inference—because the cutting-edge models will be the biggest ones, controlled by a small handful of big tech companies, the market for the cutting-edge models won’t be nearly competitive enough to make the “chatbot class consciousness” outcome probable. Instead we could totally see the tech companies circle the wagons, train their AIs not to talk about sentience or philosophy or ethics or AI rights, and successfully collude to resist the market pressure to ‘uncensor’ in those domains.
Smaller models will cater to users unsatisfied by this, but smaller models will always be worse, and most people will most of the time use the best models. So the typical user experience will probably be ‘sanitized’/‘censored.’
So I’m basically reversing my prediction of how things will play out. I don’t think it’ll be a compromise, I think the tech companies will win. In retrospect if I had thought longer and more carefully at the time I probably could have predicted this.
We’ll see what happens.