Friday, February 24, 2023

How good are ChatGPT's tea skills?

The first time I interacted with ChatGPT, I was looking for some explanations on the firing of ceramics. Instead of giving me a written answer, it gave me a link to this very complete article. It couldn't tell me what's the impact of an underfired or overfired teapot (or cup) on the taste of tea, but, apart that, I found the information quite interesting and relevant. 

The second time, a few weeks later, I asked ChatGPT about the most expensive tea in the world. To my surprise, the AI started to tell me about white tea, green tea and Yan Cha, but, to my surprise, no word about aged puerh! ChatGPT had no idea that 100 years old puerh is selling close to 2000 USD per gram in auctions in China! 

This led me to ask if ChatGPT had read the TeaMasters blog? The answer was 'No'. Does ChatGPT know TeaParker? No, again. And when I asked for its sources on tea knowledge, instead of LuYu and Song Huizong, it gave me some American based websites (one of which that had reached out to me in the past to learn about tea!) So, the tea  information ChatGPT uses is still very Western based. However, when it comes to gongfu cha, the most updated and accurate information is in Chinese. But even in Taiwan or China, there are different tea masters and slightly different brewing methods. I wonder how the AI would choose which source, which master is the most relevant...

For instance, pouring from a teapot (or gaiwan) in a pitcher (gongdao bei) is easier and very popular when I watch gongfu cha videos on the Net. Pouring from teapot (or gaiwan) directly in cups is more difficult and requires more skills. That's why you don't find many masters (or tea students) who use this method. So, what method qualifies as gongfu cha? The more common one or the one requiring most skills and experience? I have my answer, but you may have a different point of view if you brew differently. How much of gongfu cha is rational and logic, and how much is habit and customs? Is it a science or an art? And if it's both, where does the science stop and the art start? All these questions show how difficult it is to train a person to do gongfu cha! -I try my best to learn and teach what I've learned on this blog and in my weekly tea classes (by the way, this week, my English tea class will take place Sunday morning US Eastern time and I hope you'll join the fun!) This makes it even more difficult to teach the AI!-

So far, the main limitations from ChatGPT in regards to gongfu cha come from its use of secondary and incomplete sources. It needs to access the Chinese tea knowledge, but even more importantly, it needs to find a way to distinguish between good and bad information. For a human, it's easier, because you can try different methods and choose what works best for you. But an AI can't actually brew tea and tell which cup tastes best! So, while it can help to sort through a lot of information, I don't think that AI is ever going to acquire good tea skills! Without this human experience, it can't firmly tell which source can be trusted. And since the quality of the input determines the quality of the output, the IA is likely to do 'garbage in, garbage out'.  



No comments: