To not know how to contribute exactly and intelligence for that
So, o grande desafio, penso, para a pesquisa seria a questão de não saber exactly how to proceed. It is a training, a PhD training, and also there are resources and a way to contribute. Like, a modus operandi, but we can learn. Strive. Struggle. Proceed. Fail. That is key. To fail, but it is trying.
So, studying the last vídeo do curso do Karpathy Neural Networks From Zero to Hero. Depois do curso, e com meta para conclusão, depois do curso considerar esse modo de trabalhar que o Karpathy ensinou. Like, to read papers and work with code. That is key. Professor Atlas ensinou sobre o skim papers daily, let's consider that. Besides that, math and very strong code implementor. Sure, now there are the models, but there are the skills, like, to really understand. Special. But, the most important perspective is to have ideas and the skim, and reading, help to that. Focus. Action.
Cuidemos das tarefas. E aqui escrevemos o que estamos aprendendo.
Great to have the understanding for the GPT-2 and GPT-3 papers. Karpathy referencia-os. Like, it seems atrasado porque já passou, but it is there, principals. Attention Is All You Need should be first. Porque ali que se explica. Além disso, o código, pela aula do Karpathy, não se praticou muito e ficou-se de tentar like o AHR com o código, but for sure, lembrei que pensei em primeiro fechar o curso. E tem isso de GPT-2. Let's try to understand what is the Karpathy proposal for this last video. Let's focus on finishing the curso.
Great. One goal, man. Focus.
So, lembrei de um paper. Leiamos.
That paper investigating if AI agents can conduct AI research. Many limitations and interesting that I observed some of them when trying to use AI models to help with research task. Engineering ok, but the research, the questions, not yet, as the authors discusses. So, perhaps there, to this task, we can find the really challenges. Great.
Really great learning.