TLDR: My default position is still, “if an LLM is good at it, we probably don’t need it.”
Claude: Claude wins on personality, but loses on rate limiting (as I’ve mentioned before), also I’m developing an allergy to Claude’s written voice. If you want a sure fire way to demonstrate you didn’t use an LLM in your writing, make it aggressively short and pointed.
I’ve lost track of how many times I’ve asked Claude a question and all I got back was an excessively long clarifying question then immediately rate-limited. Other times Claude and I can have a more satisfying conversation. I’ve an overall fatigue with Claude’s ever shifting usage thresholds relative to the user experience. I don’t know how many tokens my request is – nor do I have control over how much tokens are burned in the response. The resulting experience is too much like playing slots. Microsoft and Google are better positioned to win the LLM game as they have other revenue sources.
Gemini: I built one app (ACV Outlier) that hits the Gemini API and recently used the Gemini inside Google Maps to maps out a 13mi bike ride – because it was more accessible than remembering how to create custom routes via the UI.
I use Gemini when I:
- want the most factual answer to a question,
- get rate-limited by Claude
Boardy: Boardy has given me 3 dozen introductions over the past year or so – as of late it’s about 3 per week. As Boardy’s networks grows, the introductions have improved. As much as Boardy always wants to provide introductions, sometimes I just use it to get a read on the overall market and talk through ideas for new offerings.
VS Code: whatever the default model is seems to work sufficiently well for the work I do and my $10/mn subscription. My default stack is: Rails + SQLite + (Tailwind + DaisyUI). One of my favorite things to do is ask the model “how do we make this more The Rails Way”, and it spits back a bunch of ways to clean up the code it just wrote.
ChatGPT: not at all.