Jeff Dean, ex-executivo do Google, abriu o jogo sobre os bastidores do Gemini em uma entrevista recente, contando desde a origem do projeto até as escolhas de design que moldaram a ferramenta. E olha, tem um detalhe curioso ali no meio: o Gemini vivia meio “devagar” em algumas áreas até que a equipe resolveu turbinar suas habilidades de programação — e foi justamente isso que destravou uma melhora surpreendente na capacidade de raciocínio do modelo.
Como nasceu o Gemini, segundo Jeff Dean
De acordo com Dean, tudo começou quando ele percebeu algo meio óbvio olhando de fora, mas que ninguém dentro do Google tinha parado pra notar: várias equipes estavam trabalhando separadamente rumo ao mesmo objetivo. Ou seja, um desperdício de esforço homem-hora que só fazia sentido se alguém juntasse todo mundo numa mesma direção.
Na entrevista, a apresentadora Dawn Song perguntou:
“So now looking back on the development of Gemini, what surprised you the most and what lessons from building Gemini do you think that you can share and can help shape the next generation of AI systems?”
Ao que Dean respondeu:
“I think Gemini is really the culmination of a few earlier research projects, both within Legacy DeepMind and Google Brain and other parts of Google Research. And at that time, we sort of realized that we were all converging to very similar kinds of directions, like training, trying to scale up the size of models we trained. We had some efforts, independent parallel efforts, on how do you make language models also be multimodal so they can understand images and so on. And I wrote a one-page memo. I’m like, this is just silly. We should just all work together. Let’s combine our people and ideas and compute resources and train one model that is multimodal from the start that brings our best people together from across multiple research organizations within Google. And I think that was really good. …And so we sort of saw a way to bring ourselves back in closer collaboration. And I think that’s been a really successful thing.”
Por que o Gemini é multimodal desde o início
Já sabíamos que o Gemini é multimodal, mas o que pouca gente sabia é que essa característica foi pensada como prioridade desde o primeiro dia de projeto. Segundo Dean, a meta sempre foi construir um modelo capaz de entender áudio, imagem, linguagem, texto e outros tipos de informação ao mesmo tempo. No entanto, o mais interessante é que foi durante o treinamento voltado para programação que o Gemini deu um salto inesperado na capacidade de raciocínio.
Nas palavras de Jeff Dean:
“I think the focus on making the model multimodal from the beginning has been really, really good. You want the model that you’re gonna use for everything to understand text and language and code and images and videos and audio and other modalities besides. So like we put a little bit of, say, LiDAR data in the training data so it at least knows that LiDAR data is a thing because that’s an important use case for, you know, further training of Gemini models. So that is important. And that’s been, I think, a success of a decision we made at the very beginning to bring that through all the current Gemini models as well. …We wanted to make the model good at lots of things. And so I think maybe our focus on making it amazing at coding was lagging a little bit and we realized that in our work to catch up on that. And I think we have good efforts underway. But I think by focusing on that, you end up with a system that is able to really do a good job of reasoning and doing other kinds of tasks that needs to sort of work its way through breaking a complicated problem down into multiple sub-pieces and so on. And so that’s a good– if you improve coding, you also tend to improve that capability in non-coding things.”
O paralelo com o aprendizado humano
Esse achado é bem interessante porque mostra como especializar um modelo de linguagem em uma única área pode gerar ganhos em outras completamente diferentes. Aliás, isso lembra bastante o jeito como nós, humanos, evoluímos em nossas próprias habilidades. Séculos atrás, o mestre samurai Miyamoto Musashi já defendia que aprender artes e ofícios fora da esgrima também fortalecia o guerreiro. Ele usava a carpintaria como exemplo: um mestre carpinteiro entende de ferramentas, planejamento, gestão de equipe e estrutura de construções — e consegue transferir esse conhecimento de volta para a arte da espada.
Assim, a explicação de Jeff Dean sobre a origem do Gemini ajuda bastante a entender o que essa IA realmente é, criando mais consciência sobre o potencial do Gemini quando encarado como um modelo multimodal completo, e não apenas como mais um gerador de texto.
O segredo de Jeff Dean para inovar de verdade
Dawn Song ainda perguntou como Dean consegue enxergar tão longe, citando projetos como MapReduce e MOE, que só foram reconhecidos como importantes anos depois de criados:
“So essentially, this is one thing that stands out about your career is that so many of these ideas, from MapReduce to MOE and so on, they actually took years before the broader community realized how important they would become. But you had the Midas touch, and you had the special, really foresight to actually build and develop these ideas and systems and so on. So I think many people in the audience would love to learn how do you do it? How do you distinguish between technologies that are genuinely foundational and those are simply fashionable? And what engineering principle that has remained surprisingly timeless throughout your career? And also as AI shifts from models to increasingly autonomous agents, what new abstractions do you think we’ll need over the next few years and decades for our future?”
Dean respondeu com bastante humildade e um conselho prático para quem quer se manter atualizado:
“I think I’ve been very lucky, maybe, as part of it. But I also think one of the things I try to do is keep track of a lot of different trends or research topics that the community is exploring. I often tell students it’s better to skim 10 papers than to read one in detail, because you kind of then get 10 points in your cloud of what might be possible, or even skim 100 abstracts. Because what you want to be able to do is connect important ideas that have not yet been connected. And sometimes when you think about a hard problem, if you have some of these things that are now kind of vaguely starting to be possible in mind, that can help you kind of shape a full solution to a problem where instead of it seeming like seven unsolvable problems, you can squint at it and say, okay, well, these five things, I know there’s sort of vague work going on that seems important and might be able to address some of the things. Then there’s two sort of additional things that I have no idea how to do. But if I work hard on them, I can imagine solving those. And sometimes that’s super important to do the things that are incremental improvements to what we’re doing now, but you want to look for the things that are you know, maybe very different ways of doing things that might be possible or might be possible soon. …I’ve done a bunch of things that have not worked out well too, so that’s also a good tip. Try lots of things that might not work. Some of them will.”
Em resumo, o depoimento de Jeff Dean mostra que por trás do design do Gemini existe uma filosofia clara: unir esforços dispersos, apostar na multimodalidade desde a concepção e aceitar que nem tudo vai dar certo de primeira. Apesar disso, foi exatamente essa mistura de colaboração, tentativa e erro que ajudou a moldar um dos modelos de IA mais relevantes do mercado atual.
Com informações de searchenginejournal.com.
