Hacker News Reader

Comment by zxienin | Hacker News Reader

> model: "text-davinci-003"

I was surprised at first, but I notice you're using GPT-3 model, not chatGPT (no API so far)

I'm not convinced that there's any substantial difference between the two.

Do you know if plain davinci-003 can continue long programming answers? I was using Chat to rewrite python scripts. It was it able to handle longish scripts easily by asking it to "continue", or simply "go on", if it ran out of room. But now they have blocked that ability since the Jan 30 update. It is quite a bit more painful now to have to break scripts up into chunks.

As far as I've seen, yes, GPT's output is quite long, and it could take anything I threw at it. If you try for your specific inputs, please let me know, I'm interested.

GPT-3's response can be as long as you are willing to pay for. The costs per request are based on how many tokens you want in the response.

No, it's limited to 4,000 or 8,000 tokens, and it will usually stop much earlier than that.

Thanks. I have managed to get chatGPT to continue again (90% of the time) by asking it

  Continue, starting from:
  last line of output

Oh I didn't think of that, thanks!

Reply on news.ycombinator.com