Here's an _ordered_ list of the most frequent ways I've been using LLMs so far.
---
Inspired by the Karpathy video by the same name: https://www.youtube.com/watch?v=EWvNQjAaOHw
---
# Search & research
ChatGPT has proven quite capable at searching, ever since they revamped their search product about 6 months ago. Since then they've also incorporated it into the new models (o4-mini, o3) and it feels highly effective.
I really enjoy the low friction question answering of ChatGPT with search. Here are some examples of how i've been using it:
- Find out how some API works and synthesize a code snippet for me
- Check my thinking on some topic (using the top model, whatever it may be at the time)
- Example: "Here's my job search plan for the next 2 months, what do you think?"
- Challenge my investment thesis
- Example: "How deep is Synopsys moat?"
# Coding
AI has gotten pretty good at coding, and I've used it to help get quite a few projects over the line.
- Full text tabs forever - Full-text browser history search
- Heyo - Mac near-instant dictation (and far superior to Apple's bult-in dictation)
- Heyo Mobile - iOS voice journaling app. Speak and you get a formatted transcript
- Vibesidian - An Obsidian plugin that can oeprate on the entire Obsidian vault, including modifying themes and plugins.
# Language
Chinese is my second language. While keeping up my language level is not a primary goal, I love being able to pop a quick thought into ChatGPT and seeing how it thinks it should be translated.
To this end I've created two custom GPTs that help me with language:
- Chinese Translator - Does what it says. There is some prompting to tell it not to be too formal.
- Chinese Fixer - I input Chinese and it tells me where it sounds unnatural.
I like the combination of both, because I can see how the LLM would approach a thought starting from English and starting from Chinese.