Our 📢 new paper📢 on active learning of treatment effects from observational data is at The Neglected Assumptions of Causal Inference workshop Join the poster session today at 12pm EST! With Panagiotis Tigas Joost van Amersfoort Andreas Kirsch 🇺🇦 Uri Shalit Yarin 📄 drive.google.com/file/d/1ABtFNw…

Come participate in the Shifts Challenge at NeurIPS 2021! Here’s a new blog post walking through the baseline in the Vehicle Motion Prediction track, with lots of explanations! Joint work w/ @YandexAI OATML_Oxford, Cambridge Speech Group link.medium.com/XErYsz0tYjb

📢 New #NeurIPS2021 paper with Panagiotis Tigas 📢 Causal-BALD: Deep Bayesian Active Learning of Outcomes to Infer Treatment-Effects from Observational Data 📄 arxiv.org/abs/2111.02275 🖥️ github.com/OATML/causal-b… With Joost van Amersfoort Andreas Kirsch 🇺🇦 Uri Shalit Yarin OATML_Oxford 🧵1/10

Feature density methods are a great alternative to ensembles for uncertainty estimation! We improve their performance by decomposing the features and fitting separate density estimators. Really fun project with @hhwpku, come see us at the BDL workshop! 👋

Milad, Sebastian Farquhar & I have discussed pruning at init for a long time. Milad’s insight gave a great new direction and with help from wonderful co-authors we finally cracked it! 🌰🥜 The code looks great too. 😎 Live long and ProsPr 🖖👽

The Dutch also once used the ounce and pound. In 1820 they moved to the metric system and REDEFINED the ounce to be 100 gram. The pound is 500g. Since 1937 they no longer have any official meaning. Butchers/grocers still use them, but they always mean the 100/500g version. 🇳🇱🇬🇧

Started an internship @DeepMind in London last week! I'm excited that this is happening in person, and have already bumped into lots of familiar faces from Twitter/Oxford. Looking forward to learning a lot from Andy Brock & team. 🥳🤖

Want to do a PhD in ML? At OATML_Oxford we work both on core ML methodology, as well as on applications of ML in lots of interesting domains. Deadline for fully funded studentships is Friday 9 December 2022 ox.ac.uk/admissions/gra…

Having so much fun working on this every day! 😊 Building the best model with friends and lots of TPUs. 🚀

👀

10,000,000 tokens achieved internally! 🚢 🚢 🚢

For all my Europe-based friends: AI Studio is now available here as well! #GoogleIO Enjoy Gemini 1.5 Pro (and Flash!) with 1M context length for free. Upload looong PDFs and ask questions! Or an entire codebase! Or a full video! aistudio.google.com aistudio.google.com

Progress on this model has been really wild! 😱 I had great fun helping out the team a bit & learned a lot from them! Check out the thread for many cool examples 👀

2 million, excluding the 97,152 token rounding error.

A very interesting opportunity to work at the intersection of data and scaling. Paul's insights have been crucial to the success of Gemini 2.0 flash (and 1.5 and 1.0 and... 👀). He makes for an excellent mentor/manager! Come help us push the frontier further 🦾.

Interested in helping us make Gemini Pro even better? The Gemini pre-training team is looking for a Research Scientist in London to push the boundaries of LLM scaling: understanding, predicting, and improving. ♊️🚀 Apply here: boards.greenhouse.io/deepmind/jobs/…

AGI will be found once we reach the bottom of all model rabbit holes. 🍻 to my fellow diggers.

This is a great look into pre-training from our friends at OpenAI. Particularly appreciated "we are a paranoid bunch" (sounds familiar 🫣).

Never will be.

In the past if you said you work in AI, people would ask ML or RL? Bayesian or frequentist? Now it’s waifu or codegen?