ChatGPT has certainly been in the news lately. It has impacted numerous aspects of life, from businesses to research and even school projects. Yet, it was not until I read an article from The Economist that I discovered Ernie, China’s leading AI chatbot launched by Baidu. Surprisingly, while ChatGPT was downloaded 1 million times in five days, it only took 19 hours for Ernie to reach the same number. This was actually despite the issues preventing the development of this chatbot. Most of the necessary chips to train the AI models are produced outside of China, and US restrictions meant that it was more difficult for Baidu to acquire these chips. After its release, people have also noticed that censorship and controversial opinions are prevalent throughout the chatbot’s responses.
Reading about these two different chatbots piqued my interest. Therefore, I decided to compare the responses of ChatGPT and Ernie to 3 questions of interest to me, namely:
1) What is Computational Linguistics?
2) What is the best university to study Computational Linguistics?
3) The same question that I asked ChatGPT in my “November Thoughts” blog - Tell me about the book “You Look Like a Thing and I Love You” by Janelle Shane.
I proceeded to test them out on ChatGPT and recorded the answers. While I do not have access to Baidu, I have some friends who do and they allowed me to use their laptop to experiment with Ernie.
For the first question, both clearly explained the main concepts that define the field of Computational Linguistics. However, Ernie focussed much more on how diverse members of the research team in the field are, ranging from computer scientists to linguists and even experts in cognitive psychology or mathematics. ChatGPT referred more to its real-life application across different industries, and gave a more detailed analysis of the different aspects of Computational Linguistics, such as Natural Language Processing (NLP), Machine Learning, Machine Translation, Speech Recognition and Sentiment Analysis.
For the second question, Ernie started by declaring that there is no clear answer, given each school has its own characteristics. It then offered some examples of Canadian universities only, such as McGill and the University of Ottawa. After the first sentence, it also never again mentioned the phrase “computational linguistics”, seeming to have forgotten that the prompt related to this field specifically and simply provided generic information such as how York University is in Toronto and the 3rd largest university in Canada. Perhaps the location of the search had an impact on the tailoring of the results. On the other hand, ChatGPT listed universities in the US as well as the UK such as Stanford, MIT, Berkeley and Cambridge, and throughout always explained why that university was strong in the field. For example, it included references to specific programs and NLP groups.
For the last question, both chatbots had similar summaries, portraying how Shane uses humour to highlight the uses of AI along with its challenges and limitations. The main difference was that while Ernie commented on Shane’s smooth writing and vivid language, ChatGPT spent more time mentioning Shane’s background as a computer scientist, as well as her blog “AI Weirdness”. It also emphasised how the title is a quote from one of her AI-generated texts, intending to provide as much context as possible.
Overall, through this experimentation, I realised that, in a way, chatbots are just like humans. They ultimately are able to convey the same general message but they all have their particular ways of formulating their thoughts, which is what makes exploring each of them so valuable.
Credit: Stock Target Advisor (Baidu, OpenAI)