By
–
Very excited about all of this, especially personalization. A model that knows you will be more helpful for you! https://t.co/6W3I9SGX7c
— Jeff Dean (@JeffDean) 2 mai 2025
Very excited about all of this, especially personalization. A model that knows you will be more helpful for you!
By
–
"Write an implementation of this paper using Tensorflow.js and visualize the training with Canvas."
— Jeff Dean (@JeffDean) 28 avril 2025
Thanks, Gemini! https://t.co/718fKWV25V
"Write an implementation of this paper using Tensorflow.js and visualize the training with Canvas." Thanks, Gemini!
By
–
Great set of slides by my Gemini colleague @FeinbergVlad on scaling considerations in large language models, addressing the fact that the classic "scaling law" work does not take into account inference cost (!), distillation, learning rate schedules, etc. https://
vladfeinberg.com/2025/04/24/gem
ini-flash-pretraining.html
…
By
–
Zaby is smarter than your average bear, thanks to Gemini! https://t.co/bEcCB0wMrm
— Jeff Dean (@JeffDean) 26 avril 2025
Zaby is smarter than your average bear, thanks to Gemini!
By
–
I should also add there was a nice paper published about the clock system that Spanner uses to achieve some of its consistency properties with low latency:
By
–
Did you know why BigTable was called BigTable? That one is a bit less subtle, I suppose..
By
–
Yes, I was one of the co-authors on the first paper, but not the second (by then I had shifted my focus towards large scale training and deployment of neural networks).
By
–
Fun fact: I helped pick the name Spanner because it would "span" all of our data centers with a single storage system. Later, my British colleagues threw a wrench into my thinking by pointing out that to them, a spanner was, well, a wrench. Leading to this meaning today:
By
–
Great to see Spanner recognized with this year's ACM SIGMOD Systems Award "for reimagining relational data management to enable serializability with external consistency at global scale". I was fortunate to be able to work on the first versions of Spanner with awesome colleagues
By
–
For (2), documents that change and where you want the updates to be reflected quickly in the retrieval system can be challenging. You usually end up building a "fresh" layer of your system that has very different data structures that can handle updates/insertions, but that is