Yann LeCun, Meta’s AI chief, speaks about general AI and scaling.

Is the key to a breakthrough in general AI more computing power and data? Yann LeCun is Chief AI Scientist of Metaon. He spoke about the recent debate on scaling that Deepmind’s Gato sparked.

Recent successes of large AI-models such as OpenAI’s DALL-E2, Google’s PaLM, and Deepmind’s Flamingo sparked debates about their importance for progress in general AI. Deepmind’s Gato, a model developed by Deepmind, has given the public debate, conducted primarily on Twitter, a boost.

Gato is a Transformer trained using a variety of data modalities including images, texts, joint moments, or proprioception. Gato processes all training data in a token-based sequence, similar to that of large language models. Gato is able to text, describe pictures, play videogames, or control robotic arms thanks to its versatile training. Deepmind evaluated the AI model using over 600 benchmarks.

Source:
https://mixed-news.com/en/general-ai-through-scaling-metas-ai-chief-yann-lecun-speaks-out/