NVIDIA lanza una IA ABIERTA — Transcript

Full transcript

1. 0:00Nvidia has just made a huge splash by
2. 0:02launching an open artificial
3. 0:03intelligence that transcribes,
4. 0:05translates, and identifies who is
5. 0:06speaking in any recording. The most
6. 0:08interesting part is that you no longer
7. 0:10need three different programs or to pay
8. 0:12any subscription fees, as it does
9. 0:14everything in a single model that anyone
10. 0:16can download. This new model is called
11. 0:18Nemotron, and it works with prompts,
12. 0:20just like when you talk to ChatGPT.
13. 0:22For example, you can give it a meeting
14. 0:24recording and ask for a transcript with
15. 0:26the name of each speaker, a translation
16. 0:28into English, or simply a summary of
17. 0:30what was said in that meeting.
18. 0:32Until now, to do this, you had to use
19. 0:34one program to transcribe, another to
20. 0:36separate the speakers, and a third to
21. 0:38translate, but with this model, you can
22. 0:41do it all at once. And here is the best
23. 0:43part: it understands 40 languages,
24. 0:45distinguishes up to eight people in a
25. 0:47one-hour audio file, and according to
26. 0:49Nvidia, it can process 100 hours of
27. 0:51audio in less than 3 minutes. However,
28. 0:53it doesn't work in real-time, only with
29. 0:55recordings, and to go that fast, you
30. 0:57need a very powerful graphics card.
31. 1:00Also, all these results were published
32. 1:02by Nvidia itself, and they have not yet
33. 1:04been verified by external users.
34. 1:07If you want to try it, comment "Nvidia"
35. 1:08and I'll send you the link right away.