VIGSIG - Amusing videos that explain LLMs, image generation, and "RAG"

[Thomas Messerschmidt]

In the three videos posted below, three different deep fake animations take you through how LLMs (ChatGPT) work, how LLMs generate images and how retrieval augmented generation (RAG) adds your data to an LLM agent. 

LLMs

https://www.youtube.com/watch?v=xU_MFS_ACrU

Image Generation

https://www.youtube.com/watch?v=k357YPzJqgA

RAG

https://www.youtube.com/watch?v=q9hlaa7DnNU

Thomas Messerschmidt

-  

Need something prototyped, built or coded? I’ve been building prototypes for companies for 15 years. I am now incorporating generative AI into products.

-

Need a great hardworking engineer? I am currently looking for a new job opportunity in robotics and/ or AI. 

[Carl Sutter]

Strangely compelling and scary how good it is.

The content seems good - wonder how much was carefully written/edited vs AI generated?  The Rap is surely a product of an LLM :-)

Using those likenesses seems like it should be illegal.  

[Chris Albertson]

No, we pick examples where the LLM is good.    Actually LLMs are very poor at most things we humans do.

[Thomas Messerschmidt]

It probably is not legal to use perfect likeness of characters, especially without proper credit and explanation. 

Yes, that is the crazy part. Someone/something is teaching you, but you don't know how accurate the information being taught is. You probably wouldn't have that issue with a course in a proper school, but when learning from a course (or any information for that matter) found online/social media, one never really knows unless one does the research to verify the data. They say that LLMs are beginning to "eat" each other. Well, as I learned in BASIC 101: GIGO (garbage in, garbage out!)

--
You received this message because you are subscribed to the Google Groups "HomeBrew Robotics Club" group.
To unsubscribe from this group and stop receiving emails from it, send an email to hbrobotics+...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/hbrobotics/89a723aa-6fcc-4aea-b6dd-52a89172ba91n%40googlegroups.com.

[Mark Johnston]

Thanks  for the deep-fake links Thomas,

The deep fakes in audio and video are  ... 'Scary' and at the same time 'Fascinating!'   

I too question the content because of possible LLM hallucinations.

Well over a couple decades ago the concept of 'fuzzy logic' I had found very interesting.   Now LLM does nothing but intense fuzzy logic.   The vectors of probability are intense in these LLMs and are fuzzy login in now real widespread usage.  Of course fuzzy logic in web queries pre-dates LLM.  Get a bit too fuzzy and they hallucinate.

There are countless examples from past crime and court shows where some recording is used as evidence.   The deep fake of voice voids usage of a great many recordings as they are now so easily faked.  You can have somebody say total trash as long as your vocal model is broad enough for general speech faking.

Thanks for sharing, I now am going to seek out other examples of deep fakes as there are likely quite a few now.

MarkJ