How I'm Using ChatGPT
Ever since I made the decision back in May to start using ChatGPT more frequently, the tool has started transforming my relationship to work in some profound ways.
This week I’d like to briefly reflect on three of the main ChatGPT use cases that have been valuable for me: research assistant, content editor, and software engineer.
I most frequently use ChatGPT as a research assistant that helps me navigate the Internet for information that I previously would have relied on Google or Wikipedia to find. Where Google gives me lists of web pages that I have to scour to find the information I’m looking for, ChatGPT gives me instant and direct answers.
Over the past few days I’ve learned that:
Egyptian hieroglyphics are considered to be one of the oldest forms of cryptography
Thomas Jefferson built a purely mechanical cryptographic machine
Internal mining functionality was removed from Bitcoin Core in the August 2016 0.13.0 release
I could have found this information using Google or Wikipedia. But being able to “chat” with a machine that can simulate intelligence pretty well and browse the internet is starting to feel more intuitive and efficient.
The solution is far from perfect. On at least one occasion, ChatGPT has “hallucinated” by directing me towards an academic source that didn’t actually exist when I tried to look it up. But so far, the upsides outweigh the downsides by far.
Another area where ChatGPT has proven useful is content editing. I’ve used the following general approach to prompting for feedback with impressive results.
Ask ChatGPT to rate my piece on a scale of 1-10 on two separate dimensions — creativity and readability
Ask for detailed editing suggestions and potential improvements
Evaluate the suggestions and incorporate, ignore, or inquire for clarification as needed.
I only share my drafts once they’ve reached the point where I would feel comfortable sharing them with an actual human. And I don’t let ChatGPT write content for me — it feels wrong and leads to bad results. But it’s a great resource for polishing my pieces in the final stages of editing and revision. One somewhat surprising issue that I’ve come across is that it isn’t actually a great spellchecker, and often seems to miss pretty obvious spelling/grammar issues.
I still prefer working with human editors whenever possible and suspect that will remain the case for a very long time. But I can’t deny the fact that it’s been very convenient to have ChatGPT’s rapid feedback as I’ve been working on this newsletter.
ChatGPT as software engineer is the use case that I’ve explored the least. It’s also the one I’m most impressed with, by far. Articulating the vision for a computer program in plain english and having the operational code for that program spit out (by a computer) feels magical.
As I mentioned a few weeks ago, a huge chunk of the grade for my Introduction to Systems Science course at Binghamton is to propose a model of what is taking place in a complex simulation. The system acts like a black box; we can observe inputs and outputs, but don’t know what’s happening inside.
An important first step is to gather large amounts of data about the state of each cell in the simulation across multiple runs. Unfortunately, no one in my group has a strong background in programming. While I could have muddled my way towards a working program on my own, ChatGPT saved me at least several hours of fussing with code.
The hard part was figuring out what I needed the script to do and clearly articulating it. Once I did the work of clearly decomposing and formulating the problem, ChatGPT could work its magic.
It’s fascinating to see how AI is impacting the art of programming and the nature of education. Taking this approach to overcoming this hurdle simply wouldn’t have been possible just a year ago!
My feelings about ChatGPT have fluctuated between giddy excitement, fearful avoidance, and now cautious acceptance. I’ve never felt such a strange mixture of complete awe and hesitant reservation about a new technology.
I’ll have much more to say about my long-term fears and hopes for the future of large language models like chatGPT in future posts.