[PLUG-TALK] Thought you all would get a kick out of these 2 screenshots

tedm at portlandia-it.com tedm at portlandia-it.com
Wed Sep 23 06:20:19 UTC 2026


Well that sucks since the fun part was seeing the output.

You can ask copilot what the back end your session is using.  They tend to
favor OpenAI

Usually my first thing is to go to bing on the edge browser and paste in "i
need a powershell script to write a script"

That triggers the co-pilot AI interface and the next thing I paste into the
entry is "clear everything I already said I have a new project"

Then "I need to write a short c program"

Generally that will get gpt-4-turbo.  When you are using the free Edge AI
client Microsoft likes to hide the exact model name they are using so if you
ask the AI what model it will say it can't tell you but if you give it a
model and say "are you this model" it will reply with a yes or no.  I'm not
sure if one of the guardrails is to never lie and that's kicking in.

I've actually coaxed quite a lot of coding effort out of the free client.
Gpt-4-turbo is just as good at writing bash and C code for Linux as it is
for writing powershell and C# for window.

The most common way I know the AI is starting to realize I'm getting actual
work done out of it and should be using the paid version is it will start
coming back with variations of "I'm too busy right now check back again"
Sometimes waiting a few minutes and typing "can you say that again" into it
will unjam it and I'll get whatever I wanted - but that's generally it.
Beyond that, more work requests will just get stonewalled.

Chrome on the other hand, when you use their free AI client in Chrome, I
generally ask it to write powershell script and that will start the AI and
then " which specific tier is currently running our conversation" will get
you the actual model in use.  Usually Gemini 3.8 Flash seems to be the most
popular one selected.  Google does not seem to have a problem giving this
info

So far though the Google AI is far better at getting more involved code out
of for free.

The funny bit I wanted to post was the gpt-4-turbo model explaining how it
couldn't help with my code since the code's purpose was attacking systems.

 I know that I'd get a far easier to use interface if I paid for a model but
I feel it's my duty to give the free Ais some real work to do.

Ted

-----Original Message-----
From: PLUG-talk <plug-talk-bounces at lists.pdxlinux.org> On Behalf Of Tomas
Kuchta
Sent: Wednesday, September 16, 2026 7:14 AM
To: Off-topic and potentially flammable discussion
<plug-talk at lists.pdxlinux.org>
Subject: Re: [PLUG-TALK] Thought you all would get a kick out of these 2
screenshots

1st - what model have you used with MS Copilot? Copilot itself is just a
router for bunch of third party models.
2nd - which Gemini model have you used?
3rd - your attachment have not made it through the email/forum pipe.

:-) T

On Wed, Sep 16, 2026, 06:59 <tedm at portlandia-it.com> wrote:

> One from IT and copilot
>
> One from Chrome and gemini
>
>
>
> I decided to brush off some of my old network testing source code from 
> a decade ago and run it through the code review on both Ais
>
>
>
> The Microsoft one blew chunks.  You can see it's output in the first 
> screen shot
>
>
>
> The Gemini one not only found a few bugs and refactored the code it 
> made helpful suggestions on how to speed things up faster and clued me 
> in to using sendmmsg()  call to allow for sending traffic even faster
>
>
>
> Ted
>
> _______________________________________________
> PLUG: https://pdxlinux.org
> PLUG-talk mailing list
> PLUG-talk at lists.pdxlinux.org
> https://lists.pdxlinux.org/mailman/listinfo/plug-talk
>
_______________________________________________
PLUG: https://pdxlinux.org
PLUG-talk mailing list
PLUG-talk at lists.pdxlinux.org
https://lists.pdxlinux.org/mailman/listinfo/plug-talk



More information about the PLUG-talk mailing list