Hey! I'm using this plugin with Whisper right now. For some reason, it works in Simulate mode only (with typing in the message). When I set the Content variable using my recognized speech in PIE, it doesn't work at all. I'm wondering if you or anyone else using this has encountered the same. Thanks!
This is so cool. Thanks for such an awesome breakdown! There really is limitless potential here. I've been wanting to make some kind of open world multiplayer game that uses ai so Im looking forward to seeing how you build more systems with this!
@@cookseyyy I'd love to know how I could make it do voice to text and then respond. But I'm worried if multiple people are talking to one npcs it'll get super garbled haha
This just kicked me back into UE and AI! This is so cool! Thank you for making this video. A quick question on it. Playing around a few hours today and I found that I've spent 5$ on OpenAI API tokens. Wondering if there is some other way to send/receive messages back-and-forth. Currently if the messages are stored in the array and sent multiple times for history and continuity it looks like it uses more and more tokens just to keep it consistent. What I am currently trying to do is having a manual, like a document about a product and ask GPT4 questions related to that document, so everytime I send the request, the document is sent again and again. Is there a way to have the document loaded in to OpenAI server, train the model according to it and then spend tokens only on requesting information back from it? Thanks!
Yeah that's a whole area to explore when using gen AI. You pay for tokens input (the context length essentially, which can get quite long when you're always feeding a whole chat in). And you also then pay for token output, essentially the length of the response. This is also related to the model you're using. There are methods for truncation and tokenization to reduce costs for bigger contexts for sure, but I'm not that experienced with that yet. With custom agents and GPTs you can pre-train them on your material which works great for PDFs and manuals like in your use case. The API is slightly different in that case though and I need to explore it further.
Cool informative video thank you! I had been experimenting trying to get a custom GPT to lead a role playing game by giving it an editable json file to use as a game state file. I never was able to get it push the story forward in a meaningful way. I'm curious if you have found a method that allows you to actually be able to play out a full narrative arc without it losing context.
Yes! This is a big challenge but there are a number of approaches. The best one I've found so far is to use a system of "Agents". So you're not using one instance of GPT to do everything but break it down into different parts such as characters and story writers. You can get them to all talk to each other to kind of share the load and get things in the right format. I've been building a system for this which I'll share soon! The other option which I've explored less is using GPT functions. They are designed to call and receive data in specific formats which should be useful, especially for working with other APIs
![How to Control Unreal Engine with Ableton Live [Tutorial, UE4, Ableton Live]](http://i.ytimg.com/vi/1vH1uEZvq98/mqdefault.jpg)
Hey! I'm using this plugin with Whisper right now. For some reason, it works in Simulate mode only (with typing in the message). When I set the Content variable using my recognized speech in PIE, it doesn't work at all. I'm wondering if you or anyone else using this has encountered the same. Thanks!
Thanks for the Tutorial, that's what i've been looking for!
thanks, but everything is very small on the screen, I can't see anything
This is so cool. Thanks for such an awesome breakdown! There really is limitless potential here. I've been wanting to make some kind of open world multiplayer game that uses ai so Im looking forward to seeing how you build more systems with this!
I'm open to suggestions!
@@cookseyyy I'd love to know how I could make it do voice to text and then respond. But I'm worried if multiple people are talking to one npcs it'll get super garbled haha
This just kicked me back into UE and AI! This is so cool! Thank you for making this video.
A quick question on it. Playing around a few hours today and I found that I've spent 5$ on OpenAI API tokens. Wondering if there is some other way to send/receive messages back-and-forth. Currently if the messages are stored in the array and sent multiple times for history and continuity it looks like it uses more and more tokens just to keep it consistent.
What I am currently trying to do is having a manual, like a document about a product and ask GPT4 questions related to that document, so everytime I send the request, the document is sent again and again.
Is there a way to have the document loaded in to OpenAI server, train the model according to it and then spend tokens only on requesting information back from it?
Thanks!
Yeah that's a whole area to explore when using gen AI. You pay for tokens input (the context length essentially, which can get quite long when you're always feeding a whole chat in). And you also then pay for token output, essentially the length of the response. This is also related to the model you're using.
There are methods for truncation and tokenization to reduce costs for bigger contexts for sure, but I'm not that experienced with that yet. With custom agents and GPTs you can pre-train them on your material which works great for PDFs and manuals like in your use case. The API is slightly different in that case though and I need to explore it further.
Great tutorial man, much appreciated!!
Glad it helped!
Cool informative video thank you! I had been experimenting trying to get a custom GPT to lead a role playing game by giving it an editable json file to use as a game state file. I never was able to get it push the story forward in a meaningful way. I'm curious if you have found a method that allows you to actually be able to play out a full narrative arc without it losing context.
Yes! This is a big challenge but there are a number of approaches. The best one I've found so far is to use a system of "Agents". So you're not using one instance of GPT to do everything but break it down into different parts such as characters and story writers. You can get them to all talk to each other to kind of share the load and get things in the right format. I've been building a system for this which I'll share soon!
The other option which I've explored less is using GPT functions. They are designed to call and receive data in specific formats which should be useful, especially for working with other APIs
Thanks a lot ! This is trully amazing !
So happy for you you've got a 4k screen ... ever considered how much of that can be read on a 2k monitor?
thank you !
Brurh um still waiting for the next one to drop 😅
So vague from 3:30 on. We’re watching tutorials for a reason dude, doesn’t help the text is illegible
He lost me at getting the API Key....