🏆A Challenge for Our Community🏆 - 🚀Inject a Demo Video in Your Prompt! 👀

[brisk panther]

Winner gets bragging rights and full community support, testing and promotion of the prompt library listing of their choice!

The Fine Print:

1. Challenge will run through April 30, 2023 12:00 p.m. CST
2. Discussion and sharing in this thread is allowed and encouraged
3. Best use of Video Injection, voted by the community wins.
4. Voting ends April 30, 2023 11 a.m. CST

👇 Watch the Video Below for Reference👇

👇 Example here:

[hollow hearth]

Interesting

@brisk panther How did you do yours? Mine, I just asked for it and it started to write the code which then became the video embed in chat instead.

If I reload the page, it reverts to just a code box.. hmm.

Anyway, gonna watch that Hermitcraft.. ttys

[hollow hearth]

For GPT-3.5, I had to say 'a webpage' instead of 'the chat'.

Anyway, I would love to know how you did yours since it just spits the video out at the top, without explaining how to embed like it did for me. I am guessing you got it to somehow return just the code first, then provide details after using prompting.

OK, it took me 3 regenerations, but I got it to just give the embedded video.

Neat concept. Too bad no one is in here to chat about it 😦

They still revert to markdown boxes when I reload, but it's cool the AI can HTML inject its own interface without trying.

They must not do input sanitization on the Assistant messages, because it's not possible to do it with your user message (that I can find).

[brisk panther]

👇 Example here:

[wise bluff]

This is what I love about Quicksilver's approach. While ChatGPT doesn't have the ability to access the internet, the chat interface is in a webbrowser, and the browser has access to the internet, and so he just imagined a way to convince the chatbot to use a functionality it's not aware of to embed internet content in the conversation.

[hollow hearth]

It's neat, but if OpenAI starts sanitizing the AI's messages for HTML injection, then it will stop working. It also only works in my live chat, if I reload it reverts to displaying the embed code. Testing with 3.5, since my GPT-4 is out of messages for the next hour.

[wise bluff]

"Attacking the problem at right angles" is a useful euphemism.

I would place money on it working the same way for Quicksilver. Still, embedding such things in a prompt that is shared can be useful for the purposes of conveying information to the user (not the AI bot), which both increases comprehension of and usefulness of the original prompt.

[brisk panther]

It will be interesting if they did scrub but not likely, as it would result in depracating it's own outputs, code etc.

[brisk panther]

This was exactly my use case for it.

[hollow hearth]

Not really, it would just have to encapsulate it in a DIV or something

[brisk panther]

I am also working on direct embed resources other than videos. It's difiicult and tedious but I'm making progress.

[hollow hearth]

When a user tries to inject HTML, the webpage sanitizes it like this

</div><h1>HTML Injection testing</h1><div>

becomes

</div><h1>HTML Injection testing</h1><div>

which still renders as text, but doesn't get included by the browser when rendering the page.

I wonder if the only reason this works is that it simulates typing out the words into the page in real-time.

[brisk panther]

@hollow hearth 👆 👆 👉 👇 👇

[hollow hearth]

QuickSilver OS prompt doesn't seem to reference any videos at all, how is it embedding a video it doesn't know about?

[brisk panther]

I haven't released this version yet

[brisk panther]

@hollow hearth Can you do a demo video of yours like I did?

[hollow hearth]

Ah, nice. So you use markdown embed. Now it's just about getting it to output it correctly and reliably without coaching.

I just burned 20 messages on trying to get an image to display. It won't get it in one-shot like the videos, I always have to correct it afterwards.

Got it, just took some phrasing

Darnit, ran out of messages again. Neat concept, but you'd need to have the media links in the prompt. You might be able to prompt it to take a link and convert it into this, but half of the time it wants to wrap the code in backticks when you get it to output as plaintext, so reliability is an issue with it.

[brisk panther]

You're getting there! Can you play the video yet?

[hollow hearth]

Oh, my hermitcraft video example was able to play

[brisk panther]

Once you figure it out, You'll be able to load in one shot!

[hollow hearth]

The other method also had playable videos, but it lost on reload

[brisk panther]

Come back when you have messages and do it again. The more people that learn this will learn something else very valuable in the process! 🧐

[hollow hearth]

OIC, i forgot to put the link in twice so it uses it for the image and for the link

I guess I could just use () for the link and put the video in the resource spot.

[prisma rampart]

I tried to create a random video, it took about 5 minutes to load and it gave an error at the end ;-;

[brisk panther]

Can you walk me through the steps you took?

[hollow hearth]

I got GPT-3.5 to do it

took 11 attempts

[brisk panther]

That's awesome! I hadn't thought of using gifs that way. Great idea!

[prisma rampart]

I already corrected it, thank you very much, I used the code that Lucid sent

I managed to create a simple interactive menu where gpt-3.5 creates a simple menu with topics such as development, music, movies and education and returns the video, unfortunately as I use chatgpt in another language I couldn't send a print here of how it turned out but now it is possible to choose the video

[brisk panther]

Thats great!

[prisma rampart]

It looked something like this, in case anyone has any ideas to improve performance

[prisma rampart]

I'm freaking out here testing the possibilities, now when choosing for example music it returns the video of various styles of music, trying to adapt it to return the title too

[brisk panther]

This is exactly what I was looking for. As a community we get better!

[prisma rampart]

corrected version, now in addition to being able to choose video styles, it returns the title of the video and the name of the channel, I believe I made all the possible improvements, in this case I just needed to add an option in the menu to be able to choose the video you want but that's it is easy

[fossil pebble]

Like this?

[prisma rampart]

can you send the code please?

[wise bluff]

My assessment is this is a session problem, and not a problem with the AI language model.

[prisma rampart]

maybe it is but I already corrected it I sent the functional code at the end returning the name of the video, the channel and everything else

[hard moon]

Have you all seen the Pokémon battle prompt from @gloomy sigil

[prisma rampart]

I will take a look

[prisma rampart]

how cool is this project

[hard moon]

Yeah we've posted a few like that on our insta and tiktok (@synapticlabs), but the prompt is all @gloomy sigil who's got a good youtube channel. Frustrating how you can't share links in here

I'm sure this could be improved as well, I don't think anyone has updated it in a while.

[prisma rampart]

great

I created a prompt that performs advanced searches in chatgpt for linkedin, facebook, instagram and twitter users, do you know if there is a way to make it return the profile image? Or do I put a static image?

@brisk panther and @hollow hearth Do you know of any way to get profile pictures or is it impossible?

[brisk panther]

As your requested tech support, I need some more details. The short answer is yes however, I need a loom video or the prompt you're using to fix it.

[prisma rampart]

can i send you private? as it is in another language and when I go to google translator it removes essential parts on the other hand in chatgpt it already runs and returns the images even accidentally

and when sending in another language here I am silenced

[brisk panther]

Yes!