Hi. thank you for your work and the library. I have used it easily. At the beginning of this video you mentioned you will show us how to add gemini support but you didnt cover it. The video was still interesting noentheless. Could you please give me a point in the right direction to configure ell for gemini models. I want to use the tools you have gave us in ell with gemini. This would be a major benefit. Thanks in advance and goodluck with the ell journey.
What are some good prerequisites for this? Looks like a great walk through, but not sure how much I would get from it without having a good understanding of recommender systems, pytorch, etc
@@MadcowDeity it’s ok Will. Stream is a big commitment. Keep it up! Also have we released the package version? I want to use ell for some of my projects for prompting instead of langchain
I think both @ell.raw and @ell.omni both should exists, I recetly did some experiments with the cohere API and they provide more functionality, such as "connectors" and "documents" to perform RAG on the backend. Therefore I think it'll be helpful to just allow users to access the raw output if they want.
Yo William! When is the stream? Also the beard is gone! I have been busy lately with some things I’m working on. I’m not as efficient and fast as you with ell. But Im slowly getting there! I hope I can catch your next stream!
@@MadcowDeity Come on let’s go! I got a semi working stream to content little tool almost done. I really want to use ell as one of the backbone of RAG and other stuff! Please come back!!! We need u!
Starred the repo! i like the minecraft parkour you added XD whats your current schedule. I'm always missing the stream and have to watch the vod after :( Also check out the continue extension for vscode. its AMAZING with codestral. check it out if you like cursor. Also whats your twitch name?
Hi, loved your approch. sometime you need the system prompt to be dynamic, {topic} in the function doc is not a bad way where you go and format the topic from the function parameters into the system prompt
1:01:01 i felt that in my bones, as I had been staring at that exact same problem for weeks. the problem is: isValidVersion(s) for some reason changes the value of idba (idba is supposed to decrypt the message in packet 6 into 24 bytes and send it back), I don't know how it does it, so what the client sends to the server has nothing to do with the for loop which puts values into. HOWEVER, there is a global variable called "want_seq", if you set that to true, it will use a different decryption algorithm (the game still runs normally) that has actually been found out by others, you can find it by googling slither protocol version 11 and search for the decodeSecret function. The game code (client code) is made confusing on purpose so people like us don't do the stuff we're trynna do.
@@MadcowDeity I remember people being obsessed with that trash - what is wrong with them, were they high on something spoiled ? Just let it rest in the trash bin of history.
Would you be open to sharing a github of the project? I don't play minecraft but I'd love to try to use this lovely framework for other similar game tasks or generalization
Very cool! Would it be possible/beneficial to combine it with some kind of image recognition model, like OpenCV? Sorry if it is a dumb question.. I'm a complete noob at this subject.
So the chatGPT model we are using GPT - 40 is actually natively able to process images. So actually, I’m not feeding it “ the block is in front of you etc”. I’m just showing it an image and saying here are some of the rules of the game, then it says output I’m going to move forward and then I process that action as a keypress I sent to Minecraft. What’s incredibles is process the world and move around in 3-D despite not ever having been trained to play video games.
Nice presentation, but don’t understand this “black canvas” passion. Or at least upload the content with higher resolution or increase the font you are using when you decide to use just 20% of the screen.
