Starbase Update, Llama 3 and Faster Data Transfers – Live and Learn #40
Welcome to this edition of Live and Learn. This time with the release of Llama 3, a new way to transfer data over fiber optic cables, Coding Adventures into text rendering, and a Starbase update by SpaceX. As always I hope you enjoy!
✨ Quote ✨
When you have no home, the whole world is your home. Be a nomadic minimalist to break dependencies on stuff. Our hunter-gatherer ancestors thrived by carrying nothing, then finding or making what they needed.
This quote is from Derek Sievers's book "How to Live". You can read the whole book for free on his website, just change the numbers at the end of the URLs to flip through the chapters. The book is a collection of awesomeness: 27 chapters of conflicting answers to the question of how to live life. It's a great (albeit short) read and I highly recommend it. No word is superfluous and every chapter hits you with wisdom.
Links
Vasa-1 by Microsoft. This paper introduces a novel way to generate talking head videos from only a single photo and a short audio sample. The results are quite impressive and this is yet another step towards deep fakes that are much easier to produce. To me, the craziest aspect is that it generalizes to languages beyond the training dataset, and can also work on cartoon figures and paintings. Watching Leonardo Da Vinci's Mona Lisa rap Paparazzi was quite the experience.
4.5 Million Times Faster Data Transfer by Aston University. One of the core problems in data centers is the maximum transfer speed over the wires between different machines. InfiniBand by Nvidia is already quite amazing, but being able to send a lot more data is nice too. Best of all, their new technology works with the same old fibre-optic cables, and can therefore be implemented in existing data centers with relative ease.
Llama 3 by Meta. Meta released their next iteration in the Llama open-source family of models. It's available in 7b, 80b, and soon 400b parameter sizes and they plan to make it multilingual, multimodal, and increase its context length too. With this, there is an open-source GPT-4 level LLM available for everyone, for free. And people are already running it locally, on Macbooks, phones, and whatnot. There was also a great episode with Mark Zuckerberg on the Dwarkesh Patel podcast, discussing a lot of the reasoning behind Meta's open-source philosophy and the future of AI research within Meta that I recommend listening to.
Starbase Update by SpaceX. SpaceX is insane and the work they are doing is truly groundbreaking. This is a video of a presentation by Elon Musk, where he introduces the rough plans of where SpaceX is going in the future (Mars ^^) while also showing off what they have accomplished so far. And it's a crazy list. 90% mass to orbit is done by SpaceX. China does 6%, and the rest of the world around 4%. They accomplished 291 Landings and 327 Launches so far. And that's only the beginning. They've deployed 5000+ Starlink satellites to orbit and their Starship is close to surviving re-entry into the atmosphere and attempting to land. It's an impressive list and SpaceX is not going to stop any time soon. When they say that they are going to make life multi-planetary, they might just succeed. And that is insanely cool.
Coding Adventure – Rendering Text by Sebastian Lague. I always thought I knew that rendering text on a computer screen was a bit of a miracle. A miracle, that when looked at closely enough, shows that human ingenuity is something absolutely crazy. This video shows just that. Sebastian Lague's coding adventures generally inspire awe in me. The way he takes apart a topic and tries to dive in and understand it for himself. To know how things are done, at least in principle, and then implement them from scratch himself, sharing his learnings in a digestible video format.
🌌 Traveling 🌌
I'm still traveling around Guadeloupe and enjoying my time here: surfing waves, reading books, meeting nice people. Life has continued to be slow in just the right amount but I want to continue to Martinique soon.
🎶 Song 🎶
Truth by Kamasi Washington
That's all for this time. I hope you found this newsletter useful, beautiful, or even both!
Have ideas for improving it? As always please let me know.
Cheers,
– Rico