China Unveils Wu Dao, a Massive AI Model Ten Times Larger Than GPT-3
The Beijing Academy of Artificial Intelligence introduces Wu Dao, a multi-modal AI system trained on 1.75 trillion parameters. Despite its massive scale and creative capabilities, experts doubt brute-force deep learning alone leads to general artificial intelligence.
The Beijing Academy of Artificial Intelligence (BAAI) unveils Wu Dao, a massive multi-modal AI system capable of generating text, audio, and images. This Chinese AI model stands out because it trains on a staggering 1.75 trillion parameters, making it ten times larger than OpenAI's GPT-3. BAAI chairman Zhang Hongjiang describes the system as a power plant for the future of AI, built on mega data and mega computing power.
Wu Dao impressively handles a variety of creative tasks, from writing poetry to producing music videos and powering virtual idols. However, specific details regarding its training datasets and practical applications remain scarce. Without this transparency, it is currently impossible to make a direct, objective comparison between Wu Dao and GPT-3.
Despite the sheer size of the model, experts express skepticism about the brute-force approach to achieving general artificial intelligence. A growing number of computer scientists believe that simply adding parameters to deep learning systems yields diminishing returns and represents a dead-end street. While Wu Dao produces incredibly convincing outputs, scaling up existing deep learning techniques does not necessarily move the needle closer to truly intelligent machines.