I support Recommendations CoreML team at @meta, and build large scale recommender systems(model & infra) for FB Reels, IG Reels and In Feed Recommendations.
Berkshire's annual report of 2024 is a worth of read: https://t.co/t3TrJPRFIY
Even though, over the course of it(1965-2024), it is far superior to S&P 500, the last 14 years(2011-2025), it is trailing behind.
I covered this week State Space Sequence Models(SSSM) over Transformers: https://t.co/Axaq1PcU9f
Also, Melange seems very interesting scheduler for heterogeneous scheduling in GPU Inference. Notion also have a write-up for their data lake re-architecture/rewrite, a lot of stuff!
Also, for the first time, a perspective that challenged all these stories that Cicero told us and probably at the least, he exaggerated a lot on the things that are advantageous to him and downplayed things that are not. Excellent book overall. fin/n
I read second time SPQR this year, I liked this book so much that, read one more time.
Mary Beard offers a fresh perspective on the history of ancient Rome, covering nearly 1,000 years from its founding to 212 CE. 1/n
Beard challenges many accepted stories about Roman history. For example, she downplays the decisiveness of Caesar's crossing of the Rubicon and questions the conventional division of emperors into "good" and "bad" rulers. 7/n
An interesting history of vector analysis: https://t.co/wRwCoWlYsE
Hamilton and his quaternions, Tait, Grassman's treatment, rising on the shoulders, finally Gibbs came and ran with it!
This week, I covered Quantization Aware Training(not as much as known as vanilla quantization) in MLOps Newsletter! https://t.co/oS7SrUUb5D
https://t.co/4I1JJgTKjc
This week, I covered ChatGPT-4o in the newsletter: https://t.co/xboeIttug3
Also, has some of my thoughts on various business models that gen-ai companies might pursue!
Very good blog post: https://t.co/z5dcJhGNdk
in debugging a benchmarking and how to do root-causing one after another validating/invalidating hypotheses on the issues/problems.