Coding with o1-preview

AI | 0 comments

o1-preview is the most recent model released by OpenAI as of this writing. Less a model than a GPT wrapped in the API that runs your query through multiple steps on OpenAI’s end instead of you doing it yourself.

I think they may have made adjustments, because it has been getting better for me.

I’ve used o1-preview extensively for a Python script to search a disk for duplicate files. It is 443 lines of actual code at this point. The script is moderately complex, allowing users to choose originals based on path, a preferred list of directories, or within a specific directory, delete duplicates, have a dry run, and more. These features require tracking a small set of variables from start through finish to insure proper processing.

Not a big deal for most humans, but something I’ve found LLMs to struggle with in the past. It’s been like they can only keep track of the changes in one or two of the most recent functions they’ve seen. o1 doesn’t seem to have this problem.

My conversation is over 8,000 lines long, and with every change it has properly tracked the context of the entire script at once. Any changes I’ve asked it to make that require updating multiple functions its handled perfectly. It’s accuracy has been very high, with only a single mistake made that I’ve spotted.

This isn’t revolutionary, but it is a major step forward, and I believe will allow much faster development of simple to moderate-complexity scripting, even for non-devs.

0 Comments

What's your $0.02?

This site uses Akismet to reduce spam. Learn how your comment data is processed.