Apart from the usual chat workflow, so far I've found two good use cases for LLMs in my personal projects.
The first one is real estate listings. A while ago, while I was looking for a house, I got really annoyed at how Weirdly Capitalized, excessively hyperbolic, grammatically incorrect, and often plain confusing listing descriptions can be. I used GPT-3.5 (well, davinci-003 back then) to take a description, remove all hypebolic speech and grammar/capitalization issues, and give me a perfectly readable result. Funnily enough, a couple months later a mobile real estate app added that very same functionality.
The second one is weather alerts. The Canadian government runs a WMS server that can be used, among other things, for retrieving weather alerts for an area. I initially wrote a script to check these and send me a Telegram notification whenever a new one showed up. Unfortunately subsequent updates to an alert don't provide... well, updates. They contain an updated version of the same message as the original alert, which means having to play Find the Differences to figure out that the rain forecast has changed from 50 to 55mm and the storm is now moving east instead of west.
I toyed with some ideas, including the fantastic dwdiff, but I wanted to have something a bit more user-friendly, so I turned to OpenAI and after much experimentation I ended up with this prompt:
System message: You are a friendly and helpful meteorologist. Your task is to provide an update regarding a weather alert. Be friendly but concise. Don't greet the user -- simply provide a weather update.
Prompt: You will be given an older weather alert, an updated version of it, and the output of running the `diff` command on them. The user has already seen the old alert. Provide an update on what has changed in the second alert. Don't mention things that remained the same between the two alerts.
### Previous alert
{previous_alert}
### Current alert
{current_alert}
### Diff
{diff}
And now if there are any updates I get actually useful messages like
The previous alert has been updated. The hazards now include nickel size hail and wind gusts up to 90 km/h. The timing has changed to this evening. Please continue to monitor alerts and forecasts for more information. Stay safe!
I'm curious if other people on HN have managed to find useful ways to integrate LLMs into their daily life.
The first one is real estate listings. A while ago, while I was looking for a house, I got really annoyed at how Weirdly Capitalized, excessively hyperbolic, grammatically incorrect, and often plain confusing listing descriptions can be. I used GPT-3.5 (well, davinci-003 back then) to take a description, remove all hypebolic speech and grammar/capitalization issues, and give me a perfectly readable result. Funnily enough, a couple months later a mobile real estate app added that very same functionality.
The second one is weather alerts. The Canadian government runs a WMS server that can be used, among other things, for retrieving weather alerts for an area. I initially wrote a script to check these and send me a Telegram notification whenever a new one showed up. Unfortunately subsequent updates to an alert don't provide... well, updates. They contain an updated version of the same message as the original alert, which means having to play Find the Differences to figure out that the rain forecast has changed from 50 to 55mm and the storm is now moving east instead of west.
I toyed with some ideas, including the fantastic dwdiff, but I wanted to have something a bit more user-friendly, so I turned to OpenAI and after much experimentation I ended up with this prompt:
And now if there are any updates I get actually useful messages like
I'm curious if other people on HN have managed to find useful ways to integrate LLMs into their daily life.