So, I’ve developed a bit of a ChatGPT addiction. It has overtaken Google and Facebook and is slowly creeping into all areas of life.
(Specifically, I use not only ChatGPT because for certain needs we have to use an analog developed by our engineers on our internal corporate network, so everything below is not only about ChatGPT, but about AI assistants in general. But for personal needs, it’s only ChatGPT for me.)
(1) Over the last six months, I’ve probably created a couple hundred Python scripts for data processing. I didn’t write any of the scripts myself (although I could; ask me again in a year or two, I might no longer be able to). To write a script for processing data, I just clearly state what I need, then closely examine the result, and if I like it, I run it. If it doesn’t work, and something needs tweaking, I tweak it myself. If it’s completely off, I ask for it to be redone. Most often, I end up with what I need. Example: read a CSV, create embeddings for all lines, cluster them, then write the results in separate files with the cluster number in the name. Or implement some complex data grouping.
I must mention bash commands separately. For example, I can’t recall how to sort lines from a file by length using command line and get the longest ones. Or I’m too lazy to remember detailed syntax for awk or jq to process something from the files through a pipe, it’s easier to ask ChatGPT.
(2) Lately, I frequently translate between Russian and English using LLMs. Rather than writing something in English myself, it’s easier to write it in Russian, get the translation, and then throw it into an email. It’s simply faster. It’s not even about the proficiency in English – of course, I could write it all myself. It’s about how much time is spent on phrasing. The argument “it’s twice as fast and clearer” beats all else. A downside—my English isn’t improving because of this.
(3) Generally, I run nearly 100% of the English texts I write through various LLMs, depending on the type of text. I ask them to correct the grammar, then copy-paste the result wherever I need—into an email or a Jira ticket. It seems I will soon have an anxiety that I sent something unreviewed. Because they always find something to correct, even if it’s just a minor thing like a missing article or a comma.
(4) When I’m too lazy to read large chunks of English text, I frequently throw them into ChatGPT and ask for a summary—sometimes in Russian. Can’t do this for work because the texts are often from clients, but if it’s really necessary, I also have access to a local LLM.
(5) I’m increasingly validating various design decisions (not visual design, but software design) through ChatGPT/LLM. I ask for criticism or additions. Often, the results make me think about what needs to be improved or what assumptions need to be added.
(6) I also use it for summarizing YouTube videos. Just download the subtitles in TXT format through Youtube subtitle downloader, throw them into an LLM, and then you can request summaries or ask questions based on them. It really helps to decide whether to watch the video or not.
What are your usage patterns?
Hi, I'm Rauf Aliev. 