Practical Use-Cases for AI/LLM
AI and learning language models get a bad wrap - with good reason. They're not wholly reliable and in many cases double-down on wrong information. I haven't had run-ins with egregious errors in results as others have shared, but I have had both Gemini and ChatGPT try to give me suggestions for rearranging anagrams and provide answers with letters that aren't in the original prompt...
That said I have had many solid use-cases for AI/LLMs. Here are two of them.
Sports Schedules
During my day job I schedule commercials for some sports broadcasts, and occasionally need to keep track of home schedules for specific teams for scheduling their digital creative. The second use isn't completely necessary, but I find it helpful so I know when to expect new audio. But, I digress.
Usually one of the producers I work with will copy the sports schedule for the junior hockey team by hand into a Word document. He'll highlight home and away games in green or red, espectively. Recently we discovered some scheduling errors, either in the date or in the time of the games. This gave me an idea: could I use a bot to parse the schedule from the hockey team's website and put the info in a table?
The answer was: yes, but also, no. It could not parse any information using just the URL, so that was a bust. It could get some of the information correct, but after a certain point it kept giving me incorrect info; either games were missing, or times were completely wrong. So I gave up on URL parsing. My next attempt was to copy the text I wanted it to parse into a table directly.
This worked! I gave it the information to work with, and had it dilute the information to a basic table. I also got it to convert home game times ahead a half-hour - because we have a pre-game show for home games. The result was a very useful table and I'm going to use it for next season.
Alt Text
I will admit this one was not my idea. Actually, it came from Lou Plummer on Mastodon: use Gemini (or probably, any LLM) to generate Alt Text for images. This is absolutely brilliant.
Here's an example of it below:
It didn't quite work out the way Lou suggested - I did give it instructions to use a particular chat only for Alt Text, but the second time I pasted an image it gave me search results instead of Alt Text. Woops. So going forward whenever I need alt text, I just upload the image with the text "alt text" and it generates various types of alt text I can use. I default to "descriptive" because in my view that's the point of alt text.
Ironically in the example above, I tried to get it to generate alt text for the screenshot but it just spit back the exact same text in the screenshot ;) So I fudged the alt text a little bit.
Anyway those are my two most recent useful, practical use-cases for an LLM. It's my opinion if we use these as tools to supplement our work, rather than replace things, there is room for these bots in the world.
Reply by email Share this post
Or if you prefer, find me on Mastodon.