The Best AI Model is the One You Use
The search for the 'best' model is a distraction. Focus on the tools that actually solve your problems today.
2024-02-13
Loading content...
The search for the 'best' model is a distraction. Focus on the tools that actually solve your problems today.
Someone asks me "which model is best?" at least once a week. I still don't have a clean answer, and I'm not sure one exists.
The leaderboard changes every few months. Whoever shipped last week from OpenAI, Anthropic, or Google is probably near the top. Smaller labs can beat them on a narrow benchmark and lose on the task you actually care about. Chasing the crown is a full-time job. I have other work.
Photographers say the best camera is the one you have with you. Same rule for AI tools. I reach for whatever is already open in my workflow, not whatever won a Twitter argument last Tuesday.
That sounds lazy until you watch someone spend three days migrating to a "better" model without shipping anything. I've done it. The model wasn't the bottleneck. I was.
I stopped treating model choice like a permanent decision. It depends on the job:
Most of the time I'm in Cursor or Claude Code on whatever default shipped that month. When something breaks, I swap models for that task and move on. The integration matters more than the name on the badge.
A hundred AI experiments will disappoint you if you pick the wrong task. Summarizing PDFs, writing boilerplate, drafting SQL against a schema you pasted in. Those work today. "Replace my judgment on architecture" does not.
The skill worth building is knowing which bucket your problem falls into before you open a chat window. Benchmarks won't teach you that. Using the thing badly for a month will.
Models keep getting better. I'd rather be someone who already knows how to prompt, review output, and fold AI into a real workflow than someone who always runs the latest weights and still treats every answer like gospel.