Ben Lee retweeted
Check out our experimental Gemini-powered AI Mode for your most difficult queries. Powered by our advanced Gemini reasoning capabilities and Google Search (what a great combo!). See 🧵 below for how to try it out!
AI Overviews are one of our most popular Search features ever. They keep getting better as models improve, and now, Gemini 2.0 is powering responses for coding, advanced math, and multimodal queries in the US, with more to come.
Today, we’re also introducing our newest Labs experiment for Search: AI Mode. You’ll get AI responses using Gemini 2.0’s advanced reasoning, thinking, multimodal capabilities + new ways to explore even more of the web.
Ben Lee retweeted
Our 1,000 Languages Initiative is an ambitious commitment to build an #ML model that supports the world’s 1,000 most spoken languages. A critical first step, the Universal Speech Model is a family of state-of-the-art speech models spanning 300+ languages →goo.gle/3yhZHHk
ALT USM’s overall training pipeline.
Ben Lee retweeted
Incredibly excited that Sundar launched the public preview of Cloud TPU v4 Pods at I/O today, with a flythrough video of a datacenter filled with them: youtu.be/nP-nMZpLM1A?t=7493! This is really three separate announcements: cloud.google.com/blog/produc…
ALT Google CEO Sundar Pichai on stage at I/O 2022, announcing a Google Cloud ML cluster with 8 TPU v4 Pods, backed by a flythrough of the Oklahoma datacenter they’re located in (from https://youtu.be/nP-nMZpLM1A?t=7493)
Ben Lee retweeted
Interested in converting the theory of "Probabilistic Machine Learning: Advanced Topics" into beautiful #JAX code? Available to work this summer? Consider applying to #GSoC. Details at
github.com/probml/pyprobml/b…
Ben Lee retweeted
Our GLaM paper with comparisons against GPT-3 is now out. We get 75% one-shot on TriviaQA vs GPT-3's 68% and show improvements on many tasks with sparse models. While using only 1/3 of compute for training and 1/2 of FLOPs for inference.
GLaM: Efficient Scaling of Language Models with Mixture-of-Experts
1.2T weight model that has better average zero and one-shot results than GPT-3 while using only ⅓ of compute for training.
Arxiv: arxiv.org/abs/2112.06905
Blog: goo.gle/3dBNLWQ
Ben Lee retweeted
Today we introduce the Generalist Language Model (GLaM), a sparsely activated model that achieves better overall performance on 29 few-shot #NaturalLanguageProcessing benchmark tasks with just a third of the training energy cost. Learn more on the blog ↓ goo.gle/3dBNLWQ
Ben Lee retweeted
Google demonstrates 63% computational efficiency, cutting edge in the industry, when training 480B parameter models on 2048 Cloud TPU-V4.
cloud.google.com/blog/topics… Link to model code: github.com/tensorflow/lingvo…
Ben Lee retweeted
Spurs players setting up an in-flight LAN for StarCraft after winning 1999 NBA season is so great—note championship trophy just chilling on the floor – at San Francisco, CA
Ben Lee retweeted
``Parallel Tacotron 2: A Non-Autoregressive Neural TTS Model with Differentiable Duration Modeling. (arXiv:2103.14574v1 [cs.SD]),'' Isaac Elias, Heiga Zen, Jonathan Shen, Yu Zhang, Jia Ye, RJ Ryan, Yonghui Wu, ift.tt/2PxQDeq
Ben Lee retweeted
I’m three seconds in and Justice League is already one of the funniest things I’ve ever seen
Ben Lee retweeted
#probowlvote Fred Warner
Ben Lee retweeted
My team in Google Brain Tokyo is hiring!
careers.google.com/jobs/resu…