I recently got access to Lex, a new GPT-3 word processor, and decided to take it for a test drive on some tea writing. The number one anxiety I and many others have about tools like Lex is whether or not they can replace me and the work I do as a writer. My stance on this until now has been: no, AI writers won’t replace me.
To test my assumption, I thought it would be interesting to ask the AI one of the most common questions I get: what does tea taste like?
For my first test, I figured I’d try out having the AI write a simple set of tasting notes. I decided to use a kamairicha I’ve been drinking lately as the test case, since the memory of my experience is still clear. I’ve added the results down below.
Although the content is entirely fictitious and produced by the AI, I still needed to provide Lex with some prompts and direction. There are also a few minor edits and improvements by me, the human. See if you can spot where I added specificity or detail to the text.
Without further adue, here’s what happens when the AI tastes tea.
It’s starting to get cool in Texas, so I wanted something comforting. To me, that usually means something roasted. I went to my tea stash and picked up my new bag of kamairicha, a Japanese pan-fried green tea, and started brewing.
The dry leaves are long and thin, and they unfurl beautifully in the hot water. The tea has a sweet, vegetal aroma.
I warmed my cup first with hot water, then dumped it out and proceeded to brew according to the package’s instructions. The tea had a pleasant fragrance and a pale green color.
The first sip was astringent, but not too much so. The tea had a vegetal taste with a slightly sweet aftertaste. I could definitely taste the grassy notes that are characteristic of kamairicha.
The second steep was even better. The astringency had faded, and the sweet aftertaste was more pronounced. I could also taste the umami flavor that is characteristic of this type of tea.
By the third steep, I noticed that the tea was starting to get a bit bitter. This is to be expected with kamairicha, so I stopped at three steeps.
Overall, this is a really nice kamairicha. It’s not too expensive, and it has a good balance of flavor elements. I’ll definitely be buying it again.
However, I did notice that the tea became quite bitter after being steeped for a longer period of time. I would recommend brewing this tea for a shorter time if you are looking for a less bitter cup.
For the price, I feel this tea is definitely worth trying. It is a high-quality tea that is both delicious and refreshing.
I’m impressed by how well the AI put together a credible set of tasting notes. However, as an experienced tea driner, I recognize how generic and frankly useless the information is. What the AI wrote applies broadly to more or less all Japanese green tea, which makes sense given that’s probably how its model was trained.
This isn’t surprising to me at all. Taste is a human sense, which makes it unique to the individual and highly subjective. However, what this demonstrates to me is what’s personal about tea is also what’s interesting to read, write, and discuss. It’s good that all tea writers keep this in mind.
In addition, I suspect some tea vendors could benefit from using tools like Lex to lay down a template for their tea descriptions and then tweak it to fit what they want. Not everyone is a writer, after all.
As for how the AI actually wrote the content, I started off with only this first piece of writing as my original input:
It’s starting to get cool in Texas, so I wanted something comforting. To me, that usually means something roasted. I went to my tea stash and picked up my new bag of kamairicha, a Japanese pan-fried green tea, and started brewing.
From there, I let the AI take over, with only minor edits and improvements and a few additional directions to generate further text (eg. “The second steep was” and “However”). I was sorely tempted to replace its description of flavors with my own, but decided not to. I wanted to let the reader decide whether or not the experience was plausible.
In conclusion, I’m not at all worried about the robots taking my writing job any time soon. The full subjective and contextual human experience remains impenetrable to the computer, which is how I expect it to stay for quite some time. However, there are some uses for tools like Lex, such as brainstorming ideas, creating outlines, and helping those who are not writers to put their thoughts in order. That said, these tools still require heavy human input to do their best work, so the robots are far from poised to take over the writing world.
After this experience, I’m more convinced than ever that writing with authenticity and humanity remains the most powerful and quintessentially human thing I can do. Tools and technology will come and go, but humans will always want to connect with each other.
A computer will never understand that.