AI + Neuro @DeepMind | Ph.D @PrincetonNeuro | ex @ptoncompmemlab. Opinions my own.
London, England
Joined March 2012
- Tweets264
- Following1.2K
- Followers507
- Likes1.9K
Luis Piloto retweeted
Introducing Gemini 2.5 Pro Experimental! 🎉
Our newest Gemini model has stellar performance across math and science benchmarks. It’s an incredible model for coding and complex reasoning, and it’s #1 on the @lmarena_ai leaderboard by a drastic 40 ELO margin. Only a handful of model releases have leaped ahead so strongly in ELO. 📈
ELO score differences map directly to win rate: e.g. a 400 ELO difference yields a ~91% win rate. Incredible that since 1.5, just a year ago, we jumped 200 ELO (300 since 1.0).
Here’s a fun example where Gemini 2.5 Pro writes code to create an animated swarm of colorful boids swimming in a rotating hexagon. 💫
Try the model for free today in AI Studio. It’s also available to Gemini Advanced users in @geminiapp. aistudio.google.com/app/prom…
Blog: goo.gle/4c3NitO
Luis Piloto retweeted
Our paper, Recursive sequence generation in crows (or Re-crow-sion, my preferred title) is out! Thanks to an awesome team (@KatharinaBrecht, @MillieJohnston_ , A_Nieder), we found that crows join primates in being able to grasp center-embedded sequences. science.org/doi/10.1126/scia…
Luis Piloto retweeted
Excellent, persistent interviewing from @Stone_SkyNews.
This is how it's done.
Luis Piloto retweeted
We're hiring! The arXiv Technical Director will lead the effort to update arXiv’s technical design and implementation, in order to meet short- and long-term strategic goals cornell.wd1.myworkdayjobs.co… @cornell_tech #OpenScience #Hiring
Luis Piloto retweeted
In new @DeepMind work led by @ToniCreswell, in collaboration with Irina Higgins, we present "Selection-Inference: Exploiting Large Language Models for Interpretable Logical Reasoning" - arxiv.org/abs/2205.09712
1/3
Luis Piloto retweeted
!!!NEW PREPRINT!!!
When do adolescents reach adult levels of executive function?
We used FOUR independent datasets (N>10,000), behavioral data from 17 distinct EF tasks, and nonlinear modeling techniques to address this and related questions. 🧵
psyarxiv.com/73yfv
Luis Piloto retweeted
Intriguingly, transformers can achieve few-shot learning (FSL) without being explicitly trained for it.
Very excited to share our new work, showing that FSL emerges in transformers only when the training data is distributed in particular ways!
arxiv.org/abs/2205.05055 🧵👇
Luis Piloto retweeted
Reinforcement learning agents can learn to cooperate, compete, and communicate in multi-agent settings. New research explores the emergence of micro-economic behaviour as agents learn to produce, consume, and trade resources in simulated worlds: dpmd.ai/Emergent-Bartering 1/
Luis Piloto retweeted
A great 🧵 here from @jalayrac. Here's a short one of my own 1/9
Luis Piloto retweeted
It is great to see the excitement about 🦩! As shown in different examples during the last few days, interacting with 🦩 has been quite fun, unique and sometimes mind blowing. However,🦩 has clear limitations as detailed in this 🧵! 1/11
Luis Piloto retweeted
Our team at DeepMind is hiring! Come and work with me and some really fun people working on human-AI interaction research. Think AI + humans + social science + alignment 🧠👩🔧
boards.greenhouse.io/deepmin…
Luis Piloto retweeted
A very powerful optical illusion. I've seen these effects before, but this one looks exceptionally potent. I wonder who created it.
Luis Piloto retweeted
Metaphors Neuroscientists Live By: frontiersin.org/articles/10.…
Luis Piloto retweeted
Recently the DeepMind Jax Ecosystem was joined by four new libraries: Mctx, KFAC-JAX, DM_AUX, and TF2JAX. See the thread 🧵 for more detail.
We hope that this fosters new research and applications in the ML community! 1/
Luis Piloto retweeted
We're open-sourcing a PyTorch implementation of SIMONe—a recent approach towards unsupervised scene understanding from @RishabhKabra & others @DeepMind.
SIMONe creates remarkable segmentations from unlabeled video! For an explanation of the paper & code: generallyintelligent.ai/open…
Luis Piloto retweeted
I’m hiring a Research Scientist to work on anymodal architectures and representation learning! If you’re interested in building systems for all the world’s data 📷⚗️🎙️📚🔬, please apply to @DeepMind's RS London role and mention you’d like to work with me boards.greenhouse.io/deepmin…
Luis Piloto retweeted
Interested in continuous time methods for ML?
Consider submitting your work to the ICML workshop on the topic: sites.google.com/view/contin…
Luis Piloto retweeted
Could machines design new tools for us? With my co-first authors @neuro_kim and @ZePoLiTaT, we are excited to announce "Physical Design using Differentiable Learned Simulators" where we solve complex, physical design problems by differentiating through GNN-based simulators. 1/9
Luis Piloto retweeted
So proud to announce that our paper on how the brain represents high-level event structure in music is now out in JoCN! We are very grateful to our reviewers that suggested additional analyses that led us to new discoveries based on our original findings.
Luis Piloto retweeted
Can we update neural models of RL to account for heterogeneous signaling of DA neurons? Read my new preprint with Ben Engelhard, @IlanaWitten, and @nathanieldaw to learn more about our Vector RPE model.
Explainer 🧵below:
A vector reward prediction error model explains dopaminergic heterogeneity biorxiv.org/cgi/content/shor… #biorxiv_neursci