High Signal Podcasts Evidence ledger
Method
Browse
← Back to evidence

Evidence receipt / prediction

Published · transcript-backed

Jimmy Wales: prediction

18 Jun 2023 Lex Fridman Podcast #385 – Jimmy Wales: Wikipedia

“My understanding is it’s improved dramatically and also because the new methods of training don’t necessarily involve identical corpuses to try to match things up, but rather reading and understanding with tokens and large language models, and then reading and understanding, and then you get a much richer … Anyway, apparently it’s quite improved, so I think that now, it is quite possible that these smaller language communities are going to say, “Oh, well finally, I can put something in an English and I can get out Zulu that I feel comfortable sharing with my community because it’s actually good enough, or I can edit it a bit here and there.”

— Jimmy Wales

Source trail

Everything needed to verify it.

Speaker
Jimmy Wales
Attribution
Verified speaker
Claim type
prediction
Recorded
18 Jun 2023
Publisher
Lex Fridman Podcast

Transcript context

…So, you start with the basic draft of the translation of articles and then build on top of that. What I used to say is machine translation for many years wasn’t much used to the community, because it just wasn’t good enough. As it’s gotten better, it’s tended to be a lot better in what we might call economically important languages, that’s because the corpus that they train on and all of that. So, to translate from English to Spanish, if you’ve tried Google Translate recently Spanish to English is what I would do, it’s pretty good. It’s actually not bad. It used to be half a joke and then for a while it was kind of like, “Well, you can get the gist of something.” And now, actually, it’s pretty good. However, we’ve got a huge Spanish community who write in native Spanish, so they’re able to use it and they find it useful, but they’re writing. But if you tried to do English to Zulu where there’s not that much investment, there’s loads of reasons to invest in English-Spanish, because they’re both huge, economically important languages. Zulu not so much. So, for those smaller languages, it was just still terrible. My understanding is it’s improved dramatically and also because the new methods of training don’t necessarily involve identical corpuses to try to match things up, but rather reading and understanding with tokens and large language models, and then reading and understanding, and then you get a much richer … Anyway, apparently it’s quite improved, so I think that now, it is quite possible that these smaller language communities are going to say, “Oh, well finally, I can put something in an English and I can get out Zulu that I feel comfortable sharing with my community because it’s actually good enough, or I can edit it a bit here and there. ” So, I think that’s huge. So, I do think that’s going to happen a lot and that’s going to accelerate, again, what will remain to most people an invisible trend, but that’s the growth in all these other languages. So, then move on to 100 years. I was starting to get scary.…

Stored transcript either side of the excerpt. The highlighted words are the published quote; the surrounding text is unedited source, never generated.

Search evidence