How I Learned to Stop Worrying and Love Language Models

My experience with ChatGPT has been nothing but incredible

I have been using ChatGPT consistently for four days to help me with my work projects. As someone peripheral to the field of AI, I had my misgivings about these large language models (LLMs), with ChatGPT arguably the most popular in the media.

I read a Wikihow article on how to use it to code, and never looked back. The first thing that really awe-struck me was when it unexpectedly wrote me a bash script, but instead of pure bash commands, it was fully written in Perl! It’s not important that the script took 8 hours to run. The thing is this code was easy to follow, neatly organized, and did what I wanted.

Today I have spent hours fine-tuning tasks that I had been putting off for months. It’s like having a personal guru. I’m no longer wasting time on StackOverflow or Biostars, in the hopes that my question has been answered! already. Even with answered questions, especially on Biostars, some of them are ancient old—not to mention the erosion of time on dependencies and platforms! Who among us has never had to download Bowtie version 1? How many of my Chrome bookmarks are old StackOverflow questions? (I tremble at the thought.)

I wrote a bash script that auto-generates a directory with a given name, complete with a README file and conda environment, designed for a generic single-cell analysis project. And that README was beautiful; it even included a section for example usages, module descriptions, and three published reference papers.

The point is that I feel confident that LLMs are not here to take my job. The results of ChatGPT are not perfect. The technology will improve. For now, I have the domain knowledge and the wisdom that the model does not. I can say that I work much more efficiently. And for that, LLMs have been a godsend.