search

apockill / clai

Command Line AI- this tool lets you call ChatGPT from any textbox on your computer
11 stars 1 forks source link

ChatGPT 3.5 or 4? #5

Open ludberg opened 1 year ago

ludberg commented 1 year ago

Is clai using ChatGPT 3.5?

When or how would it be possible to switch, or choose version?

apockill commented 1 year ago

Currently you can use the --model flag to change the model that is requested to the ChatGPT API. GPT4 isn't available (yet), but once it is I will likely change the code to support sending the screenshot to OpenAI directly, instead of attempting to run OCR to extract text.

If using shin, you could probably set SHIN_DEFAULT_COMMAND=clai --model 'whatever-gpt-4-model-name-is', so that it defaults to using that model.

ludberg commented 1 year ago

Can you make sending the screenshot an option? Sending a screenshot is somewhat worrying privacy wise

On Wed, Mar 15, 2023, 20:56 Alex Thiele @.***> wrote:

Currently you can use the --model flag to change the model that is requested to the ChatGPT API. GPT4 isn't available (yet), but once it is I will likely change the code to support sending the screenshot to OpenAI directly, instead of attempting to run OCR to extract text.

— Reply to this email directly, view it on GitHub https://github.com/apockill/clai/issues/5#issuecomment-1470751725, or unsubscribe https://github.com/notifications/unsubscribe-auth/AAYPJBQT4AQZMSXUXNNXA7LW4INGJANCNFSM6AAAAAAV4JTL3M . You are receiving this because you authored the thread.Message ID: @.***>

apockill commented 1 year ago

Yes- that's definitely in the roadmap. FYI, currently this doesn't send images anywhere. It only sends the OCR context. For example, here's the full text that is sent to OpenAI when I tried using it while writing this response.


Active Window Title: ChatGPT 3.5 or 4? · Issue #5 · apockill/clai - Google Chrome

Active Window OCR Extracted Text (RAW):
------ OCR DATA START ------
f@ (1) Master Planning Cale x | @ (1) Mind Map Pe ys)
CO CME ME cou rae ac ete
coe Bm Me oe carat]
[Berea Notifications Fa faire ate oe
ibus-bogo-pyt.
ee Ce aS et
~ Back to notifications
ludberg opened this issue yesterday -2 comments ® >
be EES /OHE © oka
Currently you can use the (=Hi6H61) flag to change the model that is requested to the ChatGPT API. GPT4 isn't available (yet),
Pee ue ee te eR a mee wR en a ORE UCR ne ey
Tetum en Pie eet (N22 ae Ue Ree Mer ere Cae
CoCr frau om
TL) ludberg 12 hours ago via email 3 re ON Aaa
UE eR eae ned
SST eee aa eT CS A
RelA AU Estea0- oY ML Ball © aia Loo
Currently you can use the ~model flag to change the model that is
requested to the ChatGPT API. GPT4 isn't available (yet), but once itis |
will likely change the code to support sending the screenshot to OpenAl
CTA URE NCR eek ti ea ota
Reply to this email directly, view it on GitHub
Seer Mesa a econ rer ace TaVAGs Stone Noval SOO ECU NINN Uo Zo VUVZ ET
Sen ee Re. ae tee CEM eR cle a
Yes: that's definitely in the roadmap. FYI, currently this doesn't send images anywhere. It only sends the OCR context. For
See ee a eee eeu
respond to this github issuel
One erae ect
OL ie omens
Assignees &
Milestone 8
eee eta ered
Ree Customize
<x Pinissue@
------ OCR DATA END ------

User Prompt:
respond to this github issue

Please answer "User Prompt" using the raw OCR text as context to the message.
ElijahLynn commented 1 year ago

Default --model is specified here https://github.com/apockill/clai/blob/c9e97bfe17b2e12f1c69d12ce1de6ee755cfe306/clai/main.py#L10