Switched from text-davinci-002 to text-davinci-003
I switched to text-davinci-003 to get better summaries, descriptions, headlines, and keywords from OpenAI, but it cost me more. I committed the data to a git repo so I could access it even if the databases got deleted. Going forward, I'll only process single posts as I explore AI capabilities. Join me as I explore the AI capabilities and the cost of processing data.
Exploring AI Capabilities: Switching from text-davinci-002 to text-davinci-003
By Michael Levin
Tuesday, April 18, 2023
After playing around with prompts and OpenAI models, I decided to re-process my entire site. To get OpenAI to write better summaries, descriptions, headlines and keywords, I switched from text-davinci-002 to text-davinci-003. It did a better job specifically with the keywords and especially the formatting of the keywords. The only problem was that it couldn’t keep the meta descriptions under 160 characters and this request ended up with a lot of sentences chopped in the middle, despite telling it specifically not to. Maybe more importantly, is it doubled my cost for the month, driving my total spend for processing about 500 pages from $30 for the first run to $40 for the recent run. Including my experimenting, I’m up to about $80 and I’m going to stop here.
Technically, I’m not stopping. I’m just not reprocessing the entire site again. I’ve committed the data to the git repo, so even if the databases get deleted I can get them back. And any time I write a new post, the process will run for the single post. It’s exciting to see the AI portions get written.