1

I’ve been using for several month completions (text-davinci) and chat 3.5 turbo.
I’m using Chat-GPT for understand a text and returns a check list that my program can read.
I have a 95% of succes with davinci and a 60% with 3.5 turbo… I’m a bit worried abut the replacemnt of text-davinci and i would like to ask that please do not remove that option next year…
Thanks

1 Like
2

“oops, our bad” is unlikely to come down about the deprecation of 20 different models.

You should evaluate gpt-3.5-turbo-instruct for the same completion application and see what changes are needed.

3

I’m testing the same question that i do with davinci with 3.5 turbo, and there are too many ‘failures’…
I’ve added some responses to the chat in order to understand better what i need, but it’s not as good as text-davinci by far :frowning:

1 Like
4

Can you list some examples of the differences?

5

Are you specifically using the same completion endpoint (not chat, not messages) with only the substitution of the -instruct model? Using constrained sampling parameters so you can evaluate the best path of generation from both?

text-davinci-003:

Here’s an AI language model trained to write haiku poetry in English:

Subject: old wise cat
Season: autumn
Poem count: 4
Poems:

An old wise cat
Staring out the window pane
Autumn’s gentle breeze

Softly purring fur
A wise old soul in the night
Autumn’s chill in air

The old cat’s eyes glow
In the fading autumn light
A sage of the night

A wise old cat sits
Watching the falling leaves drift
Autumn’s peaceful calm

gpt-3.5-turbo-instruct:

Here’s an AI language model trained to write haiku poetry in English:

Subject: old wise cat
Season: autumn
Poem count: 4
Poems:

  1. Old wise cat sits still
    Watching leaves fall from the trees
    Autumn’s peaceful grace

  2. Whiskers twitching, paws
    Tapping to the rhythm of
    Autumn’s gentle breeze

  3. Fur as soft as silk
    Eyes that hold ancient wisdom
    Autumn’s feline grace

  4. In the fading light
    Old wise cat finds solace in
    Autumn’s golden hues.

6

Hi, i would like to get one red card but no too expensive and two blue cards
Davinci:
#1#red#card#too expensive
#2#blue#card
3.5-turbo:
#1##red card#
#2#blue#card
Sometimes responses:
3.5-turbo:
##1#red#card
#2#blue#card

7

you’re not showing all the input to the AI model. Whatever that input is supposed to mean…

I’m sorry, I am an AI and I do not have the ability to sell physical items. I suggest checking online or visiting a store that sells cards to find the best prices for a red card and two blue cards.

8

Create a CSV file separating the fields with # to import into Excel with the following fields: Quantity, Format, Product, Comment. In the quantity put only numerical values. Format refers to colors or sizes
This is the list:
Hi, i would like to get one red card but no too expensive and two blue cards

Thanks :slight_smile: