Stop Saying "I used ChatGPT"
If you're just giving people ChatGPT's answers you're doing it wrong
ChatGPT and other LLMs are tools everyone is using. We need to transition from giving it all the credit to treating it like any other tool. If you are using AI correctly, you’re not copy pasting the answer from one question. You’re adding your knowledge and feedback into the system to get a result.
There’s been a lot of discourse about how everyone having access to LLMs will impact the world. One of those impacts is increasing “laziness.” In the same way calculators made math significantly easy, LLMs make a myriad of things “easier” from research, to writing, to reading. Studies have been showing that once people adopt AI they trust it and the level of review on the outputs is minimal1. A recent study from Peking University found that students using ChatGPT were more proficient in short term tasks, but had little to no impact on long-term knowledge gain. Students relied on ChatGPT to help answer questions and complete tasks, but that reduced their need to learn the material.
LLMs create an answer by predicting the most likely next word/s. The results aggregate to the mean. Here’s a whisker chart comparing the lexical diversity between press releases from 2019, 2024, and a sample of press releases I generated with ChatGPT.
What it shows is the press releases generated with ChatGPT using 3 different factors each with 100 options all sounded similar. The output from ChatGPT aggregated much closer to an average than the human generated press releases. For some tasks being similar to the average is helpful. How do you actually get value from using an LLM if you don’t want to sound average?
Add your thoughts and opinions. I often use ChatGPT or Gemini to help with research, but I always have a conversation and add my thoughts based on what it comes back with. If the result you end with from ChatGPT is something someone else can get to with 1 or 2 prompts your result is average.
Have a conversation. Most LLMs are set up as chats. Use them that way. Don’t aim for a single prompt that gets the answer perfect or good enough. Have a conversation, ask questions, clarify points.
Don’t always use ChatGPT. I intentionally don’t use ChatGPT when I write these articles. If what you’re trying to do isn’t create the average version of something, don’t use ChatGPT. Use it for research, use it to help frame something up, but don’t use it for the actual task of generation.
I’ve tried to stop saying “I used ChatGPT” because the connotation now is shifting to you didn’t do anything. The shine is starting to wear off. So keep using ChatGPT, use it correctly, but don’t give ChatGPT all the credit.
It’s human + AI. Don’t give all the credit to AI.
https://www.microsoft.com/en-us/research/uploads/prod/2022/06/Aether-Overreliance-on-AI-Review-Final-6.21.22.pdf
https://dl.acm.org/doi/fullHtml/10.1145/3584931.3606997
https://www.tandfonline.com/doi/full/10.1080/10447318.2022.2138826