> Some people can pick up languages relatively easily, and some people can’t.

I’d push back on this idea, and instead say that some people learn really well from the current tools, and some people can’t.

99.99% of the population picked up their mother tongue just like everybody else. Naturally there are differences in the population, but I would posit that if you learned a language once (or twice or three times - I know trilingual people who insist they can’t learn a foreign language because they failed Duolingo Spanish) you can learn a language again.

My bias is that I’m working on an alternative language learning tool now - one aimed at ‘language acquisition’ over ‘language learning’, or basically how can you learn a language by just talking/listening to native speakers. Aimed at those who have finished traditional courses and want to master a language, or those who have tried traditional courses and find it didn’t mentally agree with them.

> 99.99% of the population picked up their mother tongue just like everybody else.

They have - over the course of many years, completely submerged in the environment of their mother tongue, when they had an abundance of leisure time.

Not to mention their brains had more plasticity than an adult's.

The ability to learn languages is measurable.

https://en.wikipedia.org/wiki/Defense_Language_Aptitude_Batt...

I'm guessing he's suggesting the AI can be a conversation buddy anytime you want, and also point out your mistakes so you can learn from them.

imo this would be awesome and a game changer

Ah, okay yeah that would be really good. It’s hard to find time to hang out with a native speaker if you aren’t already in that country. Maybe this is already possible with ChatGPT?

spell checkers cant even get things right half of the time, not too confident AI will be able to change that

Here's an idea, but it is going to be really difficult to implement. You have lessons but at a certain point you get a chat "buddy" that only uses slightly more vocab than you know to naturally expand your knowledge. This replicates the teacher student in person discussions. You can then analyze the voice and ensure that someone has the right accent and pronunciation. Then use optimization to determine your faults and what you need to practice, optimizing memory. It isn't much different from what a human teacher would do. But I'm not aware of any of these subsystems being relatively close to human level but also not aware of much research in any of these directions. I've seen a few people try and they are just terrible.

> I plan trips to Germany every year because you really do need immersion to get to even basic fluency.

That's not necessarily true, unless consuming lots of media in the target language counts as immersion. I became fluent in English without ever traveling to an English speaking country.

Consuming a lot of media do count as immersion.

What is your native language?

Imagine ChatGPT but it uses speech-to-text to understand your speech and then it uses text-to-speech to talk back to you. Now you’re getting infinite personalized conversation practice. If you ever don’t understand what it said, hold spacebar to see the phrase in both English and Spanish with each word mapped to its translation, or just say “que?” and it’ll explain the sentence in English and/or Spanish. This seems pretty much doable today. If the speech-to-text fails to understand you, that’s immediate pronunciation feedback. May as well make it one of those boy/girlfriend chatbots too, to encourage participation and make studying more effortless.

Pls go build this, thx.

Couldn't ChatGPT? As-is, this seems almost a good prompt...