Pablo Samuel Castro
@pcastr.bsky.social
Señor swesearcher @ Google DeepMind, adjunct prof at Université de Montréal and Mila. Musician. From 🇪🇨 living in 🇨🇦.
https://psc-g.github.io/
https://psc-g.github.io/
Pinned
thrilled to finally put out the second release from my musical project "the 45s" with my buddy JS Diallo!
open.spotify.com/album/0Eb3vq...
open.spotify.com/album/0Eb3vq...
🚨The Formalism-Implementation Gap in RL research🚨
Lots of progress in RL research over last 10 years, but too much performance-driven => overfitting to benchmarks (like the ALE).
1⃣ Let's advance science of RL
2⃣ Let's be explicit about how benchmarks map to formalism
1/X
Lots of progress in RL research over last 10 years, but too much performance-driven => overfitting to benchmarks (like the ALE).
1⃣ Let's advance science of RL
2⃣ Let's be explicit about how benchmarks map to formalism
1/X
October 28, 2025 at 1:56 PM
🚨The Formalism-Implementation Gap in RL research🚨
Lots of progress in RL research over last 10 years, but too much performance-driven => overfitting to benchmarks (like the ALE).
1⃣ Let's advance science of RL
2⃣ Let's be explicit about how benchmarks map to formalism
1/X
Lots of progress in RL research over last 10 years, but too much performance-driven => overfitting to benchmarks (like the ALE).
1⃣ Let's advance science of RL
2⃣ Let's be explicit about how benchmarks map to formalism
1/X
Reposted by Pablo Samuel Castro
Proud to announce that Meta-World+ was accepted to NeurIPs, Datasets and Benchmarks! Meta-World is a common benchmark for multi-task and meta-RL research! However, it was very difficult to do effective science with Meta-World as different versions produce different results.
September 19, 2025 at 11:21 PM
Proud to announce that Meta-World+ was accepted to NeurIPs, Datasets and Benchmarks! Meta-World is a common benchmark for multi-task and meta-RL research! However, it was very difficult to do effective science with Meta-World as different versions produce different results.
🔊Simplicial Embeddings (SEMs) Improve Sample Efficiency in Actor-Critic Agents🔊
In our recent preprint we demonstrate that the use of well-structured representations (SEMs) can dramatically improve sample efficiency in RL agents.
1/X
In our recent preprint we demonstrate that the use of well-structured representations (SEMs) can dramatically improve sample efficiency in RL agents.
1/X
October 20, 2025 at 2:07 PM
🔊Simplicial Embeddings (SEMs) Improve Sample Efficiency in Actor-Critic Agents🔊
In our recent preprint we demonstrate that the use of well-structured representations (SEMs) can dramatically improve sample efficiency in RL agents.
1/X
In our recent preprint we demonstrate that the use of well-structured representations (SEMs) can dramatically improve sample efficiency in RL agents.
1/X
Reposted by Pablo Samuel Castro
Very excited to be able to talk about something I've been working on for a while now - we're working with Commonwealth Fusion Systems, IMO the leading fusion startup in the world, to take our work on AI and tokamaks and make it work at the frontier of fusion energy. deepmind.google/discover/blo...
Google DeepMind is bringing AI to the next generation of fusion energy
We’re announcing our research partnership with Commonwealth Fusion Systems (CFS) to bring clean, safe, limitless fusion energy closer to reality with our advanced AI systems. This partnership...
deepmind.google
October 16, 2025 at 1:55 PM
Very excited to be able to talk about something I've been working on for a while now - we're working with Commonwealth Fusion Systems, IMO the leading fusion startup in the world, to take our work on AI and tokamaks and make it work at the frontier of fusion energy. deepmind.google/discover/blo...
Reposted by Pablo Samuel Castro
Hello everyone 👋 Good news!
🚨 Our Game Theory & Multiagent Systems team at Google DeepMind is hiring! 🚨
.. and we have not one, but two open positions! One Research Scientist role and one Research Engineer role. 😁
Please repost and tell anyone who might be interested!
Details in thread below 👇
🚨 Our Game Theory & Multiagent Systems team at Google DeepMind is hiring! 🚨
.. and we have not one, but two open positions! One Research Scientist role and one Research Engineer role. 😁
Please repost and tell anyone who might be interested!
Details in thread below 👇
September 29, 2025 at 12:36 PM
Hello everyone 👋 Good news!
🚨 Our Game Theory & Multiagent Systems team at Google DeepMind is hiring! 🚨
.. and we have not one, but two open positions! One Research Scientist role and one Research Engineer role. 😁
Please repost and tell anyone who might be interested!
Details in thread below 👇
🚨 Our Game Theory & Multiagent Systems team at Google DeepMind is hiring! 🚨
.. and we have not one, but two open positions! One Research Scientist role and one Research Engineer role. 😁
Please repost and tell anyone who might be interested!
Details in thread below 👇
this was a really great workshop at @rl-conference.bsky.social , and i enjoyed giving a talk and moderating a panel there.
everything was recorded, so if you weren't able to join, you can watch it now, link below.
everything was recorded, so if you weren't able to join, you can watch it now, link below.
Missed our workshop at @rl-conference.bsky.social?
Watch all keynote presentations and the panel discussion on YouTube:
youtube.com/playlist?lis...
Big thanks to our awesome speakers: @togelius.bsky.social, @jmac-ai.bsky.social, Roberta Raileanu, Michael Bowling and @pcastr.bsky.social!
Watch all keynote presentations and the panel discussion on YouTube:
youtube.com/playlist?lis...
Big thanks to our awesome speakers: @togelius.bsky.social, @jmac-ai.bsky.social, Roberta Raileanu, Michael Bowling and @pcastr.bsky.social!
RLVG Sessions @ RLC 2025 - YouTube
Recorded sessions of the Reinforcement Learning and Video Games (RLVG) workshop at the Reinforcement Learning Conference (RLC), 2025. Recorded on August 5, 2025
youtube.com
September 29, 2025 at 10:25 AM
this was a really great workshop at @rl-conference.bsky.social , and i enjoyed giving a talk and moderating a panel there.
everything was recorded, so if you weren't able to join, you can watch it now, link below.
everything was recorded, so if you weren't able to join, you can watch it now, link below.
Today's wordle covered me
Wordle 1,559 2/6
🟩🟩🟨🟨⬛
🟩🟩🟩🟩🟩
Wordle 1,559 2/6
🟩🟩🟨🟨⬛
🟩🟩🟩🟩🟩
September 25, 2025 at 10:09 PM
Today's wordle covered me
Wordle 1,559 2/6
🟩🟩🟨🟨⬛
🟩🟩🟩🟩🟩
Wordle 1,559 2/6
🟩🟩🟨🟨⬛
🟩🟩🟩🟩🟩
Reposted by Pablo Samuel Castro
📢MARK YOUR CALENDARS!🗓️
The Summit of AI in Latin America (SALA, formerly RIIAA) is happening in Quito, Ecuador, March 9-12, 2026, at USFQ!
Want to attend and apply for travel grants? Do so here before October 1st:
forms.gle/PgYUQ2urGNTs...
More details in thread 👇🏾🧵
The Summit of AI in Latin America (SALA, formerly RIIAA) is happening in Quito, Ecuador, March 9-12, 2026, at USFQ!
Want to attend and apply for travel grants? Do so here before October 1st:
forms.gle/PgYUQ2urGNTs...
More details in thread 👇🏾🧵
September 3, 2025 at 10:20 AM
📢MARK YOUR CALENDARS!🗓️
The Summit of AI in Latin America (SALA, formerly RIIAA) is happening in Quito, Ecuador, March 9-12, 2026, at USFQ!
Want to attend and apply for travel grants? Do so here before October 1st:
forms.gle/PgYUQ2urGNTs...
More details in thread 👇🏾🧵
The Summit of AI in Latin America (SALA, formerly RIIAA) is happening in Quito, Ecuador, March 9-12, 2026, at USFQ!
Want to attend and apply for travel grants? Do so here before October 1st:
forms.gle/PgYUQ2urGNTs...
More details in thread 👇🏾🧵
Very happy this work got into #NeurIPS2025, it was a big effort that Reggie led, and I'm glad it was worth the work!
We've already been using Meta world+ in a lot of our research, and so should you!
Let's chat at #NeurIPS2025 !
We've already been using Meta world+ in a lot of our research, and so should you!
Let's chat at #NeurIPS2025 !
Proud to announce that Meta-World+ was accepted to NeurIPs, Datasets and Benchmarks! Meta-World is a common benchmark for multi-task and meta-RL research! However, it was very difficult to do effective science with Meta-World as different versions produce different results.
September 20, 2025 at 9:13 AM
Very happy this work got into #NeurIPS2025, it was a big effort that Reggie led, and I'm glad it was worth the work!
We've already been using Meta world+ in a lot of our research, and so should you!
Let's chat at #NeurIPS2025 !
We've already been using Meta world+ in a lot of our research, and so should you!
Let's chat at #NeurIPS2025 !
Caught my 11yo reading this, I'm so proud.
September 11, 2025 at 1:45 AM
Caught my 11yo reading this, I'm so proud.
want to help shape the future of AI in latin america? then don't miss out on @lasalaai.bsky.social next march!
the past instances of this event have been truly fantastic, and i'm super pumped for the next one next year. don't miss out and apply today!
hope to see you in quito!! ♥️🇪🇨
the past instances of this event have been truly fantastic, and i'm super pumped for the next one next year. don't miss out and apply today!
hope to see you in quito!! ♥️🇪🇨
📢MARK YOUR CALENDARS!🗓️
The Summit of AI in Latin America (SALA, formerly RIIAA) is happening in Quito, Ecuador, March 9-12, 2026, at USFQ!
Want to attend and apply for travel grants? Do so here before October 1st:
forms.gle/PgYUQ2urGNTs...
More details in thread 👇🏾🧵
The Summit of AI in Latin America (SALA, formerly RIIAA) is happening in Quito, Ecuador, March 9-12, 2026, at USFQ!
Want to attend and apply for travel grants? Do so here before October 1st:
forms.gle/PgYUQ2urGNTs...
More details in thread 👇🏾🧵
September 3, 2025 at 10:51 AM
want to help shape the future of AI in latin america? then don't miss out on @lasalaai.bsky.social next march!
the past instances of this event have been truly fantastic, and i'm super pumped for the next one next year. don't miss out and apply today!
hope to see you in quito!! ♥️🇪🇨
the past instances of this event have been truly fantastic, and i'm super pumped for the next one next year. don't miss out and apply today!
hope to see you in quito!! ♥️🇪🇨
Today's wordle was pretty minimal
Wordle 1,535 2/6
⬛🟩🟩⬛🟩
🟩🟩🟩🟩🟩
Wordle 1,535 2/6
⬛🟩🟩⬛🟩
🟩🟩🟩🟩🟩
September 1, 2025 at 2:38 PM
Today's wordle was pretty minimal
Wordle 1,535 2/6
⬛🟩🟩⬛🟩
🟩🟩🟩🟩🟩
Wordle 1,535 2/6
⬛🟩🟩⬛🟩
🟩🟩🟩🟩🟩
Software engineering isn't dead, it just smells funny.
August 25, 2025 at 12:55 AM
Software engineering isn't dead, it just smells funny.
Yesterday I did a 28.4k trail run and 1300m climb with some good friends (& colleagues). It was absolutely incredible! This was my first trail run, but now I think I'm hooked!
#trailrunpostconference
#trailrunpostconference
August 11, 2025 at 2:05 PM
Yesterday I did a 28.4k trail run and 1300m climb with some good friends (& colleagues). It was absolutely incredible! This was my first trail run, but now I think I'm hooked!
#trailrunpostconference
#trailrunpostconference
Once again, I leave @rl-conference.bsky.social convinced that it's the best conference out there for RL researchers, such a great week!
It was also so great getting the opportunity to see so many of my students shine while presenting and discussing their research!
It was also so great getting the opportunity to see so many of my students shine while presenting and discussing their research!
August 9, 2025 at 1:23 PM
Once again, I leave @rl-conference.bsky.social convinced that it's the best conference out there for RL researchers, such a great week!
It was also so great getting the opportunity to see so many of my students shine while presenting and discussing their research!
It was also so great getting the opportunity to see so many of my students shine while presenting and discussing their research!
Did the last #runconference @rl-conference.bsky.social this morning, fun as always.
Until the next time (NeurIPS?)!
Until the next time (NeurIPS?)!
August 9, 2025 at 5:06 AM
Did the last #runconference @rl-conference.bsky.social this morning, fun as always.
Until the next time (NeurIPS?)!
Until the next time (NeurIPS?)!
Was great listening to Turing award winner Rich Sutton tell us about his vision for the future of RL (and AI) @rl-conference.bsky.social !
August 9, 2025 at 3:37 AM
Was great listening to Turing award winner Rich Sutton tell us about his vision for the future of RL (and AI) @rl-conference.bsky.social !
Great talk by Peter Dayan @rl-conference.bsky.social exploring the "open marriage" between natural and engineered reinforcement learning.
August 8, 2025 at 6:43 PM
Great talk by Peter Dayan @rl-conference.bsky.social exploring the "open marriage" between natural and engineered reinforcement learning.
Great #runconference @rl-conference.bsky.social today (even with a little rain), join for the last one tomorrow morning, 6:30am, meet at Garneau Lamp!
August 8, 2025 at 1:09 AM
Great #runconference @rl-conference.bsky.social today (even with a little rain), join for the last one tomorrow morning, 6:30am, meet at Garneau Lamp!
today (thu) @rl-conference.bsky.social come see our paper "Optimistic critics can empower small actors"
olya and dhruv will present the oral at 11:45am in the deep RL track (CCIS 1-430)
and we'll present the poster (poster#8) 3-5:45pm
come chat with us about asymmetric actor-critics!
olya and dhruv will present the oral at 11:45am in the deep RL track (CCIS 1-430)
and we'll present the poster (poster#8) 3-5:45pm
come chat with us about asymmetric actor-critics!
📢optimistic critics can empower small actors📢
new @rl-conference.bsky.social paper led by my students @olyamhina.bsky.social & dhruv!
we study actor-critic agents w/ asymmetric setups and find that, while smaller actors have advantages, they can degrade performance and result in overfit critics
1/
new @rl-conference.bsky.social paper led by my students @olyamhina.bsky.social & dhruv!
we study actor-critic agents w/ asymmetric setups and find that, while smaller actors have advantages, they can degrade performance and result in overfit critics
1/
August 7, 2025 at 2:06 PM
today (thu) @rl-conference.bsky.social come see our paper "Optimistic critics can empower small actors"
olya and dhruv will present the oral at 11:45am in the deep RL track (CCIS 1-430)
and we'll present the poster (poster#8) 3-5:45pm
come chat with us about asymmetric actor-critics!
olya and dhruv will present the oral at 11:45am in the deep RL track (CCIS 1-430)
and we'll present the poster (poster#8) 3-5:45pm
come chat with us about asymmetric actor-critics!
Very honoured that our paper was granted an outstanding paper award for scientific understanding in RL during the @rl-conference.bsky.social banquet!
@reggiemclean.bsky.social and Evangelos deserve the bulk of the credit, so congrats to them and all the other awards winners!
@reggiemclean.bsky.social and Evangelos deserve the bulk of the credit, so congrats to them and all the other awards winners!
August 7, 2025 at 4:28 AM
Very honoured that our paper was granted an outstanding paper award for scientific understanding in RL during the @rl-conference.bsky.social banquet!
@reggiemclean.bsky.social and Evangelos deserve the bulk of the credit, so congrats to them and all the other awards winners!
@reggiemclean.bsky.social and Evangelos deserve the bulk of the credit, so congrats to them and all the other awards winners!
Super thought-provoking talk by Dale Schuurmans @rl-conference.bsky.social on LLMs and computation, and why value-based RL doesn't (or can't?) work for post-training.
August 6, 2025 at 9:07 PM
Super thought-provoking talk by Dale Schuurmans @rl-conference.bsky.social on LLMs and computation, and why value-based RL doesn't (or can't?) work for post-training.
Second #runconference today went well, join us tomorrow same place (Garneau Lamp), same time (6:30am)!
August 6, 2025 at 8:03 PM
Second #runconference today went well, join us tomorrow same place (Garneau Lamp), same time (6:30am)!
today (wed) @rl-conference.bsky.social come see our paper "Multi-Task Reinforcement Learning Enables Parameter Scaling"
@reggiemclean.bsky.social will give oral at 10:20am in CCIS 1-160
and we'll present the poster 3-5:45pm
come chat!
rlj.cs.umass.edu/2025/papers/...
bsky.app/profile/regg...
@reggiemclean.bsky.social will give oral at 10:20am in CCIS 1-160
and we'll present the poster 3-5:45pm
come chat!
rlj.cs.umass.edu/2025/papers/...
bsky.app/profile/regg...
August 6, 2025 at 2:15 PM
today (wed) @rl-conference.bsky.social come see our paper "Multi-Task Reinforcement Learning Enables Parameter Scaling"
@reggiemclean.bsky.social will give oral at 10:20am in CCIS 1-160
and we'll present the poster 3-5:45pm
come chat!
rlj.cs.umass.edu/2025/papers/...
bsky.app/profile/regg...
@reggiemclean.bsky.social will give oral at 10:20am in CCIS 1-160
and we'll present the poster 3-5:45pm
come chat!
rlj.cs.umass.edu/2025/papers/...
bsky.app/profile/regg...