Categoria
🎥
CortometraggiTrascrizione
00:00I've been waiting for this for 40 years
00:07Denise has the world's largest collection of dolphin vocalizations
00:13I'm a research scientist at Google DeepMind
00:17Dolphin Gemma is the first LLM trained to try to understand dolphin language
00:24Dolphin Gemma will input sounds
00:27Once a dolphin starts doing a vocalization like a whistle
00:31It can try to complete the end of it
00:32When you're doing a Google search, right, it's finishing your sentence, right?
00:36Dolphin Gemma has Denise's data
00:39And sort of encapsulates a lot of the knowledge and experience she has in it
00:42But it's also small enough we can train it with more data as we get it
00:45We can actually keep on fine-tuning the model as we go
00:49And hopefully get better and better understanding of what the dolphins are producing
00:54We do not know if animals have words
00:58Dolphins can recognize themselves in the mirrors
01:00They use tools
01:01So they're smart
01:03But language is still the last barrier
01:05So feeding dolphin sounds into an AI model like dolphin Gemma
01:10Will give us a really good look at if there are patterns, subtleties that humans can't pick out
01:17If dolphins have language, then they probably also have culture
01:21You're going to understand what priorities they have
01:24What do they talk about?
01:25The goal would be to someday speak dolphin
01:28And we're really trying to crack the code
01:30That means Dad
01:35Save us
01:35Aunty
01:38And we're trying to crack the code
01:39Untilonna
01:39We're really trying to get people
01:40And we're trying to learn how to move
01:41Entire
01:42And this is what we work
01:42We're trying to see
01:44You
01:44We're trying to see
01:48What we create
01:49Yes
01:50What we're...
01:51We're trying to find a human
01:52We're cara
01:52We're trying to look
01:54Ones
01:55We're trying to Cecil
01:56We're trying to see
01:57We're trying to face
01:58When we're trying to move
01:58As 1980