Will Dabney
@wwdabney
Followers
1K
Following
12K
Media
5
Statuses
148
Research scientist at DeepMind. On the critical path to AGI. Also, a persistent optimist.
Seattle, WA
Joined May 2015
OpenReview seems to be essentially down? I guess I’m not the only one submitting last minute NeurIPS reviews.
2
0
3
Love this response.
now wouldn't that be something. .
0
0
7
The moment all of my agents have trained for has finally arrived. Real world impact here we come!.
Porsche Cayenne attempts to escape the swamp that no other vehicle has been able to exit.
2
0
24
Pulling this over as research advice (for myself):. Feedback beats planning. It’s a lesson I have to relearn now and then because of the allure of an intuitive hypothesis.
I have never seen it expressed exactly like that, but I wholeheartedly endorse it:. Feedback beats planning. My plea at Meta was “No grand plans, follow the gradient of user value”.
0
0
16
RT @clarelyle: 📣📣 My team at Google DeepMind is hiring a student researcher for summer/fall 2025 in Seattle! If you're a PhD student intere….
0
74
0
Some fantastic news about a very well deserved award! Congratulations Andy and Rich, today the entire field is celebrating you and with you.
0
0
35
RT @neuro_kim: Very pleased to share our recent work, in which we use LLMs for automated discovery of interpretable models of animal behavi….
0
10
0
Winding down from a fantastic turnout for “Normalization and effective learning rates in reinforcement learning” led by @clarelyle . Zeyu Zheng and James Martens at the poster explaining how we are solving plasticity loss at #NeurIPS2024 ! . Paper:
1
1
20
I need to say something not related to AI. My neighborhood in Seattle has lost power and driving home just now I barely missed getting T-boned. It forced me to admit, if the worst happens, I would be completely on my own. No emergency services would come. Seattle is broken.
2
0
9
RT @JesseFarebro: Introducing the Distributional Successor Measure (DSM): a model of the range of possible futures an agent faces. As a dis….
0
25
0
Undeniably impressive rate of progress.
Introducing Sora, our text-to-video model. Sora can create videos of up to 60 seconds featuring highly detailed scenes, complex camera motion, and multiple characters with vibrant emotions. Prompt: “Beautiful, snowy
0
0
7
Couldn’t agree more with this, and it is an area where we have seriously lost our way recently.
This is one of the hidden secrets of scalable research labs: creating template experimental codebases that allow fast testing of wild research ideas. You need your experiment, data, compute, and model abstractions just right 1/.
0
0
10
I really didn’t have that one on my bingo card. Wow! So curious about what is going on behind the scenes.
0
0
11
This work was a collaboration with Zeyu Zheng, @nikishin_evg, Bernardo Avila Pires, and Razvan Pascanu. A great group thinking deeply about this intriguing topic.
0
0
1
What happens when neural networks lose plasticity?.What properties cause plasticity loss? .How can we mitigate plasticity loss?. @clarelyle's work on "Understanding Plasticity in Neural Networks" thoroughly investigates these questions, providing both insight and advice.
1
1
16
Today at #ICML2023, Charline is presenting her work diving into details of representation learning with and for reinforcement learning. Her analysis can also yield keen insights into several of the more empirical papers being presented this week as well. Poster #202, 11am HST.
Today at #ICML2023: How shall we (pre)train principled representations in RL at scale from a limited computational budget?. We analyze and derive new auxiliary tasks to do so!🌟. First in-person #ICML, first time in Hawaii, last PhD paper! 🌴🌺.
0
0
20