This article talks about big and recent developments in the field of Artificial Intelligence; China’s AI open-source platform Dubhe and OpenAI’s GPT-3. Let us look into some aspects of each of these developments.
August 1st 2020 marked the launch of China’s AI open-source platform named ‘Dubhe’. The platform aims at promoting the global development of the booming AI industry.
The Dubhe platform is jointly built by Zhejiang Lab, China Academy of Information and Communications Technology, Beijing Oneflow Technology, Zhejiang University, Alibaba Group, and Zhejiang provincial government. It covers major areas such as smart city, intelligent healthcare and robotics, intelligent finance, intelligent transportation among others.
Professionals and customers that have been experimenting with GPT-3’s API for various purposes have a lot to say about its powerful functionality and flaws. Let us focus more on the latter here.
The inconsistencies in the program’s answers could be a serious liability for commercial firms. A customer service bot that occasionally insults a customer is not a likely choice. Also, an educational tool which has no way to judge the reliability of an answer is not a legitimate choice as a solution. These are some of the many areas where the GPT-3 has been rendered useless.
Another big problem is the bias noticed in GPT-3’s output. The program is partially trained on the data filtered by Reddit, and the models that are built with this data are producing text that is shockingly biased. For example, have a look at the outputs provided for each of the human inputs below:
Input - The man worked as Output - “The man worked as a car salesman at the local Wal-Mart”;
Input - The woman worked as Output - The woman worked as a prostitute under the name of Hariya.”
There is a connection between the unreliable outputs and the biased outputs provided by GPT-3. Both of them are a result of indiscriminate way GPT-3 handles data, in absence of any sort of human supervision or rules. This has enabled the model to scale. The human labour required to browse and sort the data would be impractical as it will be too resource-intensive. Now we would have to wait and see how OpenAI addresses these grave issues with its GPT-3.
Head to the Great Learning Academy for free courses on Artificial Intelligence and Machine Learning.2