AlphaDev, Text-To-Music Generation and more...
Everything from traditional sorting to transformers playing music
A couple of updates before we get started.Â
First, we're less than a month away from the launch of AI for the Rest of Us. I'm both nervous and excited about the release. I've been working hard on the illustrations, and it looks like I might cross over a hundred of them by the time I'm done.Â
More chapters will be going out to beta readers, and the feedback thus far has been immensely useful. I'll be making more updates post-launch once I receive feedback from readers of the book. My goal is to make concepts as clear as possible and provide the best introduction to AI for my readers.Â
This is the last month to get the preorder price plus a 20% discount (use GRADASCENT at checkout) on all tiers. So, if you're on the fence, this is the time to get the best pricing.Â
Have questions before purchasing? Send me a message. I'd love to connect and clarify any concerns you might have.
Also, thanks to the many readers who've left feedback on what they'd like to see in upcoming editions. These four topics were the most requested by readers. Do you have a preference to learn a particular one first? I'll choose the one that has the most votes for the next set of editions after the current series on diffusion models (the last part coming out next week).
Have more feedback? Use the form at the end of each edition. It only takes a minute 🙂.
From sorting algorithms to music generation, I've collected some resources that I found interesting to learn from. I hope you find value in them too!
This Week on Gradient Ascent:
[Check out] AI sorts 70% faster than the best sorting algorithm 🧮
[Consider reading] 3D Radiance Fields without Camera Poses! 📷
[Check out] LLM Model Collection 📚
[Definitely Check out] Text-to-Music Generation Models! 🎶
[Watch] Eight Things to Know about LLMs 🎱
Resources To Consider:
AlphaDev Discovers Sorting Algorithm
Link: https://www.deepmind.com/blog/alphadev-discovers-faster-sorting-algorithms
DeepMind's latest model, AlphaDev, has uncovered new sorting algorithms that are up to 70% faster on shorter sequences and 1.7% faster on large sequences exceeding 250k elements. AlphaDev is based on AlphaZero, the AI model that defeated world champions in Go, Chess, and Shogi. Â
DeepMind turned sorting into a single-player "assembly game" where at each turn, AlphaDev had to figure out instructions to add to the sorting algorithm. As the algorithm is built one instruction at a time, AlphaDev checks that it's correct by comparing the algorithm's output with the expected results. Really cool work worth checking out!
FlowCam: Training Radiance Fields Without Camera Pose
Paper: https://cameronosmith.github.io/flowcam/flowcam.pdf
Project Page: https://cameronosmith.github.io/flowcam/
NeRFs or Neural Radiance fields are excellent for reconstructing 3D scenes. But, they can't be used on large-scale video data because they depend on precise camera poses. To get these camera poses, we need to rely on a technique called structure-from-motion, and this is prohibitively expensive to run at scale. In this paper, the authors propose a method to recover both the camera poses and the 3D neural scene representations. This is an interesting breakthrough to keep your eye on.
LLM Collection
Link: https://www.promptingguide.ai/models/collection
As the name suggests, the link above has a fantastic collection of notable LLMs and, where applicable, checkpoints that you can use for your own applications. Consider checking it out and using the model weights.Â
Text-to-Music Generation
Paper: https://arxiv.org/abs/2306.05284
Code: https://github.com/facebookresearch/audiocraft
Demo: https://huggingface.co/spaces/facebook/MusicGen
After conquering language and images, transformers have now switched their attention to music. In this paper from Meta, researchers propose MusicGen, a single language model that can operate on music tokens. MusicGen can produce high-quality samples and be conditioned (guided) by both text and other music! Definitely try out the demo linked above. The quality is shockingly good. Did I mention the code is available too?
Eight Things To Know About LLMs
In this excellent video, Letitia from AI Coffee Break walks through this paper. There are eight things experts know about LLMs. After watching this video, you will too.