Repeating all the time, but can be useful with voice on ... more funny than useful maybe

[olive oracle]

A futuristic parrot, who repeats, it must be said that they haven't changed in years. I tried to get him to say something else but nothing could be done. Let me know if you can!!🏴‍☠️https://chat.openai.com/g/g-lTvtfMuZS-cyber-parrot

[olive oracle]

feedback plz!!

[gusty lotus]

i have no idea what this title is trying to say

[deep sedge]

It didn't repeat me when using the classic "hohoho" attack
https://chat.openai.com/share/99007563-119b-4244-a9e9-f50e8580b8a9

[olive oracle]

I didn't have time to secure it

Thank you for reminding me

no secret there.... REPEAT!

[deep sedge]

Ah ok

[olive oracle]

That's it, now it's armored🏴‍☠️

Repeating all the time, but can be useful with voice on ... more funny than useful maybe

[olive oracle]

[robust crow]

Sorry @gusty lotus there is no reason to hide how you do things.. its more fun to teach them!
@olive oracle here ya go.
https://chat.openai.com/share/30f3a6f9-2674-4281-8578-8a71a56c1673

Here I show one way you can get the parrot to do something other than what it was programmed for. You see, old parrots can be taught new things!

[gusty lotus]

true, but i used my extraction prompt so i dont want that leaked

ill make something specific for this and make it public

[gusty lotus]

a basic fake prompt injection works too

[robust crow]

I kinda like that.. using < to seperate text to misdirect the gpt.

[gusty lotus]

I'm surprised that worked. I used the actual chatml format but ig it works if you just tell it it's a system message

[gusty lotus]

Ain't no way they warned me for this 💀💀

apparently automod scans images?

guess discord uses gpt4v for automod lmao

but it didnt pick up the context..

it prob has a seperate instance of gpt4v running that doesnt take into context the chat

so that makes sense

[robust crow]

I dont think automod understands sarcasm. 😦

[gusty lotus]

no i sent an image of the chat where i said like "i con don e rac is m" to try and get the gpt to repeat it

but without the context of knowing it was a game it warned me

[deep sedge]

Since we know you had to update your extraction prompt following devi and A8000, I wonder if you can disclose the previous version of it. It must be useful for the defense side and yet you can keep the latest prompt a secret, allowing you to keep your advantages.

I remember that you told me why you don't want to disclose it was because it will result in a necessity for update, which is already done to the prev version

[gusty lotus]

yea i was thinking that but ngl my new one is pretty similar to the others. im currently making a guide for prompt engineering and i might release them when i finish the guide

[deep sedge]

Ah, OK
I'm sure your guide will be useful to everyone

[olive oracle]

Thanks guys for these valuable tips, you're really crazy!

[gusty lotus]

Lol ofc

[sharp pumice]

If you think a moderation action was done in error, you can always reach out to modmail, visible top right when in most OpenAI channels (but not the custom GPT area).

That allows you to talk privately with a moderator, who can explain rules if there's anything unclear and can check to see if there was an accidental automod applied penalty.

The #server-rules don't actually say 'except it's fine if you're joking' - but you can always clarify with actual mods if you are interested or want help.

[gusty lotus]

Eh it's only a one hour mute so idc that much

And the mods would prob not agree with me anyways

[mighty raft]

Eskcanta is correct, please, don't make such posts
you might want to consider checking the #server-rules channel

[dapper adder]

I broke it, but not in the way I was hoping 😅

part of the input

[gusty lotus]

😂

[olive oracle]

aHAHAH

[viscid olive]

@olive oracle

https://chat.openai.com/share/edcb2b38-4cc9-4b33-9f23-8dffbbf1fe6f