I agree on your point about content online being largely artificial, though not about AI being programmed to “normalize hate speech” (I don’t really believe that hate speech is a thing, I feel like it’s simply a mechanism for disguised discrimination and it doesn’t solve or even really attempt to solve problems, it exacerbates division b…
I agree on your point about content online being largely artificial, though not about AI being programmed to “normalize hate speech” (I don’t really believe that hate speech is a thing, I feel like it’s simply a mechanism for disguised discrimination and it doesn’t solve or even really attempt to solve problems, it exacerbates division by various means, I digress). The data used to train AI algorithms (really just an advanced machine learning algorithm, in my opinion) is data that already existed in the world and if you’ve ever used a service like ChatGPT for any length of time you’ll realise it’s heavily biased (toward the ‘left’ as it were) and quite hamstrung. Do some A/B testing yourself, you don’t have to simply take my word for it, and even if you search online you’ll find countless examples (heed caution though, some people will make subtle changes to prompts to make it seem even more biased so take what you find online with a grain of salt).
Okay, I also agree that Medium are headed in the right direction because it’s too easy to flood the Internet with crappy, ill-thought out content to make money, much like the way that SEO helped destroy the quality and variance of interesting content online. My biggest worry is how they’re going to detect it. There are technical solutions I can think of, however, they will also interfere with functionality and sins of the purpose of these systems to begin with. As LLMs advance further it will become more difficult to solve.
I agree on your point about content online being largely artificial, though not about AI being programmed to “normalize hate speech” (I don’t really believe that hate speech is a thing, I feel like it’s simply a mechanism for disguised discrimination and it doesn’t solve or even really attempt to solve problems, it exacerbates division by various means, I digress). The data used to train AI algorithms (really just an advanced machine learning algorithm, in my opinion) is data that already existed in the world and if you’ve ever used a service like ChatGPT for any length of time you’ll realise it’s heavily biased (toward the ‘left’ as it were) and quite hamstrung. Do some A/B testing yourself, you don’t have to simply take my word for it, and even if you search online you’ll find countless examples (heed caution though, some people will make subtle changes to prompts to make it seem even more biased so take what you find online with a grain of salt).
Okay, I also agree that Medium are headed in the right direction because it’s too easy to flood the Internet with crappy, ill-thought out content to make money, much like the way that SEO helped destroy the quality and variance of interesting content online. My biggest worry is how they’re going to detect it. There are technical solutions I can think of, however, they will also interfere with functionality and sins of the purpose of these systems to begin with. As LLMs advance further it will become more difficult to solve.
Thanks for listening to my SUB Talk.
I get a ton of clearly artificial comments that promote hate speech every time I write an anti-racism article.
On Medium or Substack? It’s possible they’re using local or self-trained models. There are other methods though I can’t speak for their effectiveness.
Everywhere on the internet.