AIs simply do not sound like people. They can fool me for a little while, but eventually you realize it's just a bot. But that's when I'm a native speaker. How will I be able to detect this when I struggle with the language?
I can't help but think it's dangerous to give a non native speaker access to this, and lead them to believe they're learning to speak like a real person. They aren't -- they're learning to speak like GPT-2.
The chief sin of AI is: this model is sufficiently close to a human. (How'd that work out for the resume filter? Or the AI responsible for setting bail?)
The worse sin is teaching someone to emulate what the AI does. You take whatever biases the AI has, and encode it into a living, breathing person.
> AIs simply do not sound like people. They can fool me for a little while, but eventually you realize it's just a bot.
Why would a language learner be concerned about that? If I thought it was a real person, I'd become alarmed, because that's not what it was presented to me as.
> You take whatever biases the AI has, and encode it into a living, breathing person.
Biases towards certain adjectives? I don't even understand this comment. It could lead you into speaking weird Spanish if it speaks weird Spanish, but I imagine that would be something they were trying to improve if they noticed it. And growing up in a weird place or having a weird family could also result in you speaking weird Spanish. You still speak Spanish, which is what language learners are looking for, and if you want to sound like a particular class of Spanish speaker, you would hang out with them and consume their media.
I mean, my own English is pretty bad in my opinion, but it's a dream level for an aspiring language learner. Practicing their English with me would be useful, even if my speaking is somehow "biased." Are you telling me that language practice with an AI is worthless, or even a sin?
I really don't understand a bit of this comment. So much so that I'm suspicious that it was generated by a model.
who said it's going to replace human conversation? You can go to Italki and speaking to a person in any language with as little as 10$. It's a dream to all language learners to have safe practice environment. Your other arguments are the as same as those made with textbook which is rigid and "biased" by the authors.
AIs simply do not sound like people. They can fool me for a little while, but eventually you realize it's just a bot. But that's when I'm a native speaker. How will I be able to detect this when I struggle with the language?
I can't help but think it's dangerous to give a non native speaker access to this, and lead them to believe they're learning to speak like a real person. They aren't -- they're learning to speak like GPT-2.
The chief sin of AI is: this model is sufficiently close to a human. (How'd that work out for the resume filter? Or the AI responsible for setting bail?)
The worse sin is teaching someone to emulate what the AI does. You take whatever biases the AI has, and encode it into a living, breathing person.