Why LLMs Will Not Replace Google

Good Morning AI Runners 🏃‍♂️

Here's what we've got for you today:

• Co-pilot for Music: Forego the process of learning piano with AI

• Scaling Vision Transformers to 22 Billion Parameters

• Why LLMs Will Not Replace Google

Co-pilot for Music

This is a "physical" Co-pilot that goes beyond the use of OpenAI APIs to create a co-pilot for various tasks. The technology is demonstrated through "Piano Genie," which basically takes low-quality piano playing and turns it into High-quality music on the fly. The way it works is by using an autoencoder that maps piano notes to simple controller buttons, and then turns them back into piano notes. The team behind it had trained the system on over 1,400 piano performances from real people. "Piano Genie" was trained on 1400 piano performances and was developed in 2018 before the advent of large language models and co-pilots, which is why it used LSTM. Oh, and in case you're curious, LSTM stands for Long Short-Term Memory, which is a type of neural network that's really good at processing sequential data, like music or speech. It's kinda like your brain's memory, but for machines. Click the link above for a demo.

If you're into piano, try the web demo here: 

Scaling Vision Transformers to 22 Billion Parameters  

This research paper presented ViT-22B, the currently largest vision transformer model at 22 billion parameters.

We have been using these really huge language models with over 100 billion parameters to help with natural language processing, but they haven't been as successful with image and video modeling. But now, The authors (so many couldn't keep count) of this paper have come up with a way to train a Vision Transformer with 22 billion parameters that might be able to work well on a bunch of different tasks like balancing fairness and performance, aligning with how humans see things, and being more robust. This could be a big deal for things like self-driving cars and robots.

Why LLMs Will Not Replace Google

"LLMs as they exist today will never replace Google Search. Why not? In short, because today’s LLMs make stuff up."

Pic of the day:

That's it from RunTheAI for today.

THANK YOU FOR READING AND SEE YOU TOMORROW, SUBSCRIBE TO STAY UPDATED!

P.S. if you made it this far, hit “reply” and tell me what you think of today's newsletter...what’d you love? What was boring?