Vai al lettorePassa al contenuto principale
  • 6 ore fa
Trascrizione
00:00I've been waiting for this for 40 years.
00:05Denise has the world's largest collection of dolphin vocalizations.
00:10I'm a research scientist at Google DeepMind.
00:13Dolphin Gemma is the first LLM trained to try to understand dolphin language.
00:20Dolphin Gemma will input sounds.
00:24Once a dolphin starts doing a vocalization like a whistle,
00:27it can try to complete the end of it.
00:28When you're doing a Google search, right, it's finishing your sentence, right?
00:32Dolphin Gemma has Denise's data and sort of encapsulates a lot of the knowledge and experience
00:37she has in it.
00:38But it's also small enough we can train it with more data as we get it.
00:42We can actually keep on fine-tuning the model as we go and hopefully get better and better
00:47understanding of what the dolphins are producing.
00:50We do not know if animals have words.
00:54Dolphins can recognize themselves in the mirrors.
00:56They use tools.
00:57So they're smart, but language is still the last barrier.
01:02So feeding dolphin sounds into an AI model like Dolphin Gemma will give us a really good
01:08look at if there are patterns, subtleties that humans can't pick out.
01:13If dolphins have language, then they probably also have culture.
01:17You're going to understand what priorities they have.
01:20What do they talk about?
01:21The goal would be to someday speak dolphin and we're really trying to crack the code.
01:26They are having people withository.
01:29Let's go.
01:29Let's go.
Commenti

Consigliato