Show HN: Compare new OpenAI model vs. old on HN Summaries (news.jiggy.ai)

4 points by wskish 3y ago ↗ HN

The new OpenAI "gpt-3.5-turbo" model is cheap but how does it perform? Check out its summarization style compared to the 10x more expensive "text-davinci-003" model on summaries of Hacker News stories.

3 comments

[ 3.0 ms ] story [ 16.6 ms ] thread

wskish 3y ago ↗

This is with no prompt change except for putting the entire previous prompt into a single "user" role message.

williamcotton 3y ago ↗

Yeah, I’ve noticed that the quality is the same or better.

And it is so much faster as well!

O__________O 3y ago ↗

Are these the only prompts used with the models:

https://github.com/jiggy-ai/hn_summary/blob/master/src/summa...

Also, how did you decide on those specific prompts?

Lastly, is it accurate that the text being summarized is truncated instead of for example streaming it in chunks to create a summary of summaries for longer text. If so, why even process the articles that require truncation?