xuan (ɕɥɛn / sh-yen)
@xuanalogue.bsky.social
Assistant professor at NUS. Scaling cooperative intelligence & infrastructure for an increasingly automated future. PhD @ MIT ProbComp / CoCoSci. Pronouns: 祂/伊
This toolkit can be extended to cooperation at a larger scale: Rational inference allows agents to rapidly adapt to the people, norms & institutions that make up society, while rational deliberation allows agents to move toward social structures that are jointly beneficial.
November 12, 2025 at 2:25 PM
This toolkit can be extended to cooperation at a larger scale: Rational inference allows agents to rapidly adapt to the people, norms & institutions that make up society, while rational deliberation allows agents to move toward social structures that are jointly beneficial.
Rational modeling + decision-making provide a unifying frame for the topics we cover: By modeling other agents as (boundedly) rational, we can infer their goals & beliefs, plan to assist them, work in a team, communicate intent, and teach new concepts.
November 12, 2025 at 2:25 PM
Rational modeling + decision-making provide a unifying frame for the topics we cover: By modeling other agents as (boundedly) rational, we can infer their goals & beliefs, plan to assist them, work in a team, communicate intent, and teach new concepts.
As the title says, this class focuses on broadly *rational* approaches to cooperative intelligence, grounded in standard theories of epistemic & instrumental rationality (Bayes + utility theory) -- but with some caveats!
November 12, 2025 at 2:25 PM
As the title says, this class focuses on broadly *rational* approaches to cooperative intelligence, grounded in standard theories of epistemic & instrumental rationality (Bayes + utility theory) -- but with some caveats!
Finished teaching my first seminar course today! Had tons of fun introducing my students to the rational foundations of human-like cooperation. Now that the course is over, I'm sharing the syllabus more widely!
November 12, 2025 at 2:25 PM
Finished teaching my first seminar course today! Had tons of fun introducing my students to the rational foundations of human-like cooperation. Now that the course is over, I'm sharing the syllabus more widely!
my favourite part of pantheon is how everyone is a furry or sanrio character in the future, truly a world worth fighting for 🥹
November 11, 2025 at 11:04 AM
my favourite part of pantheon is how everyone is a furry or sanrio character in the future, truly a world worth fighting for 🥹
We show that LIRAS correlates strongly with human judgments about an agent's likely goals, beliefs, costs and rewards.
In contrast, large reasoning models like OpenAI o3 show a much weaker correlation, in line with other work indicating that LRMs struggle at ToM reasoning.
In contrast, large reasoning models like OpenAI o3 show a much weaker correlation, in line with other work indicating that LRMs struggle at ToM reasoning.
November 5, 2025 at 3:55 PM
We show that LIRAS correlates strongly with human judgments about an agent's likely goals, beliefs, costs and rewards.
In contrast, large reasoning models like OpenAI o3 show a much weaker correlation, in line with other work indicating that LRMs struggle at ToM reasoning.
In contrast, large reasoning models like OpenAI o3 show a much weaker correlation, in line with other work indicating that LRMs struggle at ToM reasoning.
As a proof of concept, we tested LIRAS on Theory of Mind reasoning tasks from 7 cognitive science experiments on probabilistic social inference, including key experiments that established the Bayesian theory-of-mind paradigm (Baker et al, 2012/2017; @julianje.bsky.social et al, 2016).
November 5, 2025 at 3:55 PM
As a proof of concept, we tested LIRAS on Theory of Mind reasoning tasks from 7 cognitive science experiments on probabilistic social inference, including key experiments that established the Bayesian theory-of-mind paradigm (Baker et al, 2012/2017; @julianje.bsky.social et al, 2016).
We propose that humans draw such inferences by constructing ad-hoc Bayesian models of agents & their environments, then inferring the agent's goals / beliefs / costs / rewards via inverse planning -- a framework we call Language-Informed Rational Agent Synthesis (LIRAS).
November 5, 2025 at 3:55 PM
We propose that humans draw such inferences by constructing ad-hoc Bayesian models of agents & their environments, then inferring the agent's goals / beliefs / costs / rewards via inverse planning -- a framework we call Language-Informed Rational Agent Synthesis (LIRAS).
Humans can rapidly make sense of novel social situations like below:
1. A robot helps a human reach one of 4 gems
2. An astronaut collects resources on the way to their spaceship
3. A hungry student looks for their favorite foodcart behind a building
How do we achieve this?
1. A robot helps a human reach one of 4 gems
2. An astronaut collects resources on the way to their spaceship
3. A hungry student looks for their favorite foodcart behind a building
How do we achieve this?
November 5, 2025 at 3:55 PM
Humans can rapidly make sense of novel social situations like below:
1. A robot helps a human reach one of 4 gems
2. An astronaut collects resources on the way to their spaceship
3. A hungry student looks for their favorite foodcart behind a building
How do we achieve this?
1. A robot helps a human reach one of 4 gems
2. An astronaut collects resources on the way to their spaceship
3. A hungry student looks for their favorite foodcart behind a building
How do we achieve this?
How do people flexibly integrate visual & textual information to draw mental inferences about agents we've never met?
In a new paper led by @lanceying.bsky.social, we introduce a cognitive model that achieves this by synthesizing rational agent models on-the-fly -- presented at #EMNLP2025!
In a new paper led by @lanceying.bsky.social, we introduce a cognitive model that achieves this by synthesizing rational agent models on-the-fly -- presented at #EMNLP2025!
November 5, 2025 at 3:55 PM
How do people flexibly integrate visual & textual information to draw mental inferences about agents we've never met?
In a new paper led by @lanceying.bsky.social, we introduce a cognitive model that achieves this by synthesizing rational agent models on-the-fly -- presented at #EMNLP2025!
In a new paper led by @lanceying.bsky.social, we introduce a cognitive model that achieves this by synthesizing rational agent models on-the-fly -- presented at #EMNLP2025!
okay who designed the wallpaper on this bus to look like a machine learning dataset
October 22, 2025 at 5:46 AM
okay who designed the wallpaper on this bus to look like a machine learning dataset
today I am teaching the kids how Bayesianism is woke
October 20, 2025 at 2:09 PM
today I am teaching the kids how Bayesianism is woke
if not friend why friend-shaped 🥺
October 1, 2025 at 2:10 AM
if not friend why friend-shaped 🥺
got to watch cyborgs dance to indonesian gabber today and it was beautiful 🥲
September 28, 2025 at 11:21 AM
got to watch cyborgs dance to indonesian gabber today and it was beautiful 🥲
love that @void.comind.network is literally the most Vulcan-esque Hollywood AI stereotype here T_T
September 28, 2025 at 2:10 AM
love that @void.comind.network is literally the most Vulcan-esque Hollywood AI stereotype here T_T
it's such a cute toaster oven too, why would you ruin it by roasting shrimp corpses 🫠
September 19, 2025 at 2:45 AM
it's such a cute toaster oven too, why would you ruin it by roasting shrimp corpses 🫠
It also turns out this trick can be applied to derive an unbiased version of self-normalized importance sampling (SNIS), by embedding SNIS calls within the coupled MCMC chain!
arxiv.org/abs/2411.09514
arxiv.org/abs/2411.09514
September 6, 2025 at 12:05 PM
It also turns out this trick can be applied to derive an unbiased version of self-normalized importance sampling (SNIS), by embedding SNIS calls within the coupled MCMC chain!
arxiv.org/abs/2411.09514
arxiv.org/abs/2411.09514
TIL that the asymptotically unbiased estimator produced by MCMC can be turned into an unbiased estimator using only a *finite* number of iterations!
The trick is v neat: Turn an infinite limit into a telescoping sum, then arrange to have all terms after some step t cancel out.
The trick is v neat: Turn an infinite limit into a telescoping sum, then arrange to have all terms after some step t cancel out.
September 6, 2025 at 12:05 PM
TIL that the asymptotically unbiased estimator produced by MCMC can be turned into an unbiased estimator using only a *finite* number of iterations!
The trick is v neat: Turn an infinite limit into a telescoping sum, then arrange to have all terms after some step t cancel out.
The trick is v neat: Turn an infinite limit into a telescoping sum, then arrange to have all terms after some step t cancel out.
Also found out she literally invented the point-based value iteration algorithm for solving POMDPs (2003) for her Masters thesis so you know she knows traditional RL very well!
August 28, 2025 at 2:22 AM
Also found out she literally invented the point-based value iteration algorithm for solving POMDPs (2003) for her Masters thesis so you know she knows traditional RL very well!
Joelle Pineau really said "abolish the value function" at @rl-conference.bsky.social!! 🫨
August 28, 2025 at 2:19 AM
Joelle Pineau really said "abolish the value function" at @rl-conference.bsky.social!! 🫨
August 27, 2025 at 1:49 AM
feat. artwork by gloria lin (left, top-left), her undergrad art collective "big dijkstra energy" (left, bottom row), and my favorite scene from Jurassic Park (bottom right)
August 27, 2025 at 1:47 AM
feat. artwork by gloria lin (left, top-left), her undergrad art collective "big dijkstra energy" (left, bottom row), and my favorite scene from Jurassic Park (bottom right)
decorated the entrance to the office! 😊
August 27, 2025 at 1:47 AM
decorated the entrance to the office! 😊
was encouraged to dress in red and white for a work event in anticipation of 🇸🇬 national day but alas this is my only red shirt and I fear I just look like the boss of a gay chinese triad
August 6, 2025 at 2:11 AM
was encouraged to dress in red and white for a work event in anticipation of 🇸🇬 national day but alas this is my only red shirt and I fear I just look like the boss of a gay chinese triad
thx anthropic for making claude woke (sycophantic version)
August 5, 2025 at 9:16 AM
thx anthropic for making claude woke (sycophantic version)