“It’s a exciting progress – says David Bau, a professor at Northeastern University who deals with AI, regarding the new OpenAI research –. As an industry, we need to learn to understand and examine these large patterns much better“.
According to Bau, the main innovation of the OpenAI team is to show a most efficient way to set up a small neural network which can be used to understand the components of a larger system. However, the teacher adds that to become more reliable the technique must be perfected: “There is still much work to be done to use these methods and obtain fully comprehensible explanations“, He says.
Bau is part of a US government-funded project called the National deep inference fabric, which will make cloud computing resources available to academic researchers to enable them to probe particularly powerful AI models. “We need to understand how ensure that scientists who do not work in large companies can also do this work“, comments Woof.
In their article, the OpenAI researchers recognize that it is necessary to continue working to improve the new method, hoping however that it can lead to the development of Practical methods for controlling AI models. “We hope that one day interpretability will provide us with new ways to think about the safety and robustness of models, and that it will significantly increase our confidence in powerful AI models by offering strong guarantees about their behavior“, the document reads.
This article originally appeared on Wired US
