Learnings From Pet Projects
- I tried to build a voice cloning from English to German. I wanted to translate a podcast to an “easy” B1 German, here’s the result. I was using f5-tts-mlx, and tried to train a duration prediction model so the speed of the speech is consistent. You can make a version of this in English way quicker - read Voice Cloning.
- For this I spent some time searching where to find a cheap GPU. Spent lots of time trying to make pytorch work for intel cloud, but without a success. They really should invest more resources into a software stack and integrate their support with accelerate and lightning! In the end I followed this guide, Cost-Effective Cloud Platforms for GPU Machine Learning
Links
- Interesting discussion about a founder that did an IPO and know doesn’t know what to do anymore.
- There’s a foundation that invests a lot into biology. Very curious about their work with AI!
- Nice podcast on the ideas generation, a good discussion on how the view of brainstorming and applying it changed over time.
- Another podcast about workaholics and people that choose to “live to work” instead of “working to live”, their view on the world.
- And the last podcast on the philosophy of “Demons”, that also discusses western liberalism and “Good times create weak men” part. Had lots of connecting dots to a recent US election and Berlin downfall.
- Very good review of the book about SpaceX, especially relevant now, when people have a polarized opinion on Elon Musk. Also watch this video of a Starship blowing up, looks like something from a movie!
- A primer on machine learning in cryo-electron microscopy (cryo-EM). Very interesting to learn about this field, now even more interested into figuring out if AI can help to speed up the research there more.
- Very relatable article on the broken processes. I haven’t seen things to be that bad, but I heard so many similar stories personally. Always reminds me of this book, though I read a good thought that maybe these bullshit jobs are actually luxury jobs when companies are too rich to compete properly?
And lastly a joke I heard:
What linux users and cyclists have in common? They both worry about drivers