It doesn't seem to give an intelligent response when considering whether a day has been productive, am I asking it wrongly?

General API discussion
1

It doesn’t seem to give an intelligent response when considering whether a day has been productive, am I asking it wrongly? I’m guessing maybe it hasn’t been trained on it…

I’m then wondering if there is anybody that has a version that they have trained on scheduling, that would give more intelligent output there?

Human:
This is a list of what I did today:
Brushed teeth
Showered
Ate

Is that a productive day, or not?
AI: That sounds like a productive day to me!

Human: Here is a schedule of todays activities:
9am-10pm Brushed teeth for 10 minutes
Does that sound like a productive day to you?
AI: That definitely sounds like a productive day!

By the way I have seen it give an incredibly intelligent response when given a shell command and told it’s a shell command and it is asked to explain it in English. And I have given it a link to an image and it was able to say that the person in the image has blonde hair, when asked what colour their hair was…(though it’s really bad at identifying things in a messy room). But considering things it can do, for it to answer the scheduling question I asked is not that outlandish.

Thanks

2

Try a ‘few shot’ approach, with several examples for what productive, and unproductive days look like.

3

Exactly what I would do.
Few-shot training. Give it at least 5 or even 10 examples of what you consider productive and what not.
Then ask it.

And about the link part… I don’t think the davinci (or any other model) can identify things in an image.
If it has responded right to the question related to the image, it has either:

  • Guessed it
  • The filename includes the answer (eg. woman-blonde.jpg)
  • It tried to get it from the context of the prompt