(Originally Posted in February 2023)
OpenAI’s ChatGPT has been receiving a lot of attention lately. Microsoft even invested some USD 10 billion in this AI tool. After watching a YouTube video about OpenAI’s free ChatGPT preview, I quickly created an OpenAI account and started testing it out, suspecting that they may soon put up a paywall.
ChatGPT is quite peculiar. On the one hand, ChatGPT appears to be a one-of-a-kind coding tool; on the other hand, meaningful thought seems to be absent when it comes to noncoding-related matters.
As a coding tool, I have these thoughts. I would gaze at its answers to my prompts as it generated code for various tasks I gave it. It was blazing fast and effortless. One small quirk about this tool, which I think OpenAI will eventually iron out, is that sometimes it cannot tweak its own code.
I wanted to write code that would plot graphs with certain specifications. I also wanted the graph itself to depend on user inputs. After communicating my instructions to ChatGPT and having a short back-and-forth conversation between myself and this AI (with me giving feedback), it generated the code (which worked).
The quirk occurred when I asked it to tweak the program to allow for more flexibility with user input. This should have been relatively easy. However, it generated long and convoluted code that didn’t work.
It seems likely that setbacks such as the above quirk are only temporary; after all, this is only a research preview, and an upgraded version may be coming out sometime this year. This tool seems to be more limited with more complex coding tasks. But, after seeing its current progress, my impression is that such issues will also be largely resolved over time.
Later on, I wanted to test ChatGPT for its ability to self-reflect. I asked ChatGPT to write a review for itself. After a long pause—perhaps I should dismiss this pause as network lag—it printed a series of facts in paragraph form. But the paragraph looked as if it were written by someone clumsily trying to promote him/herself. My impression was that it was reading off a list of factoids. I also asked it to generate a longer response in its review. And it did more or less the same, but with more paragraphs and, I think, more facts.
I then asked for its thoughts on Nietzsche’s philosophy. In its several-paragraph response, it wrote about what could be the major perspectives of his philosophy and known implications of his work. It seems to read like a nice Wikipedia article. Specifically, ChatGPT scraped data from the internet to use as training data, and it seems likely that it simply collected thoughts from others on this philosopher. Perhaps this machine’s response could serve as a study tool.
Overall, looking at this tool makes me feel I am much older than I am and that the world is shifting beneath my feet. It would be interesting to see what tools such as ChatGPT evolve into in the future.