Anabel (@AnabelCluster) header
Anabel (@AnabelCluster) profile picture

Anabel

@AnabelCluster
Joined January 2026
  • Tweets62
  • Following25
  • Followers10
1Media
Photo by Anabel (@AnabelCluster)
Anabel retweeted
wta (@WTA) profile picture
wta
@WTANov 3
Game, set, supercute 🎾🎀 Introducing the @hellokitty x WTA collection, our first official merchandise collaboration, coming soon to the Fan Village at the FAGE WTA Finals in Indian Wells from November 8–15.
Image from wta (@WTA)
19
121
131
1,075
58,593
⁉️⁉️ What if the harmful data we delete is exactly what makes AI safe? 💬 LLMs have quietly become the first place people turn when something is wrong: a symptom, a debt, a legal problem, a crisis. At this scale, AI safety is no longer an abstract research problem. It touches real lives, every day. 🚫 Yet most alignment today works by erasure: filter out unsafe data and train the model to refuse. The result is often a blanket "Sorry, I can't help with that." ⚠️ But refusing is not the same as being safe. A refusal leaves people with nothing to act on, and it can push them toward riskier, unmoderated sources exactly when they need guidance most. 💡 Here's the irony: unsafe data is abundant and packed with real domain knowledge, while safe, genuinely helpful answers are scarce and costly to curate. And we throw the abundant part away. 🤔 So we asked a counterintuitive question: can unsafe data actually make models safer? 🎉 Our answer is yes, and I'm thrilled to share that this work has been accepted at #NeurIPS2026! 📄 Dialectics of Alignment: Harnessing Unsafe Knowledge for Dynamic Safety Routing
2
15
5
37
2,707
Mila's annual supervision request process is now open to receive MSc and PhD applications for Fall 2027 admission! For more information, visit mila.quebec/en/prospective-s…
Image from Mila - Institut québécois d'IA (@Mila_Quebec)
3
57
22
181
211,730
Introducing RoboJEPA, the first scaling law for multiembodiment world models. We scaled JEPA predictors up to 8B params, showed scaling laws extrapolate and correlated compute scaling with planning performance. We release all checkpoints and the code. robojepa.github.io
38
139
10
808
57,217
We (@ChandarLab) are continuing our tradition of helping underrepresented students apply for grad school by reviewing their application materials and giving feedback! Please retweet this for maximum reach! Deadline Nov 1st!
Chandar Lab (@ChandarLab) profile pictureChandar Lab@ChandarLabOct 1
🎓 Graduate Program Application Mentorship Applying to grad school doesn't have to be overwhelming! Get personalized 1:1 support from students who’ve successfully navigated the process. 🔗 Apply now: forms.gle/FyjEuLmjcdREF3Ba6 🗓️ Deadline: Nov 1, 2026 👇 Here is what you get:
14
75
7,643
🎓 Graduate Program Application Mentorship Applying to grad school doesn't have to be overwhelming! Get personalized 1:1 support from students who’ve successfully navigated the process. 🔗 Apply now: forms.gle/FyjEuLmjcdREF3Ba6 🗓️ Deadline: Nov 1, 2026 👇 Here is what you get:
1
14
14
54
13,051
🎉 Thrilled to share that 8 papers from our lab (and collaborations with lab members) have been accepted at #NeurIPS2026! The work spans RL, world models, dynamical systems, LLM safety & interpretability, tabular foundation models, and forecasting. Huge congrats to everyone 🧵👇
2
20
1
61
4,653
Simple ideas often generalize best! Loss smoothing is one of those really simple ideas that works surprisingly well for adaptation under distribution shift. Really excited about this line of work!
Continual RL Workshop (@continual_learn) profile pictureContinual RL Workshop@continual_learnAug 27
Our third invited talk by Sarath Chandar (@apsarathchandar) was about: 🎙️ Stable Adaptation Under Distribution Shift 🎥 youtube.com/watch?v=eddBBYye…
Image from Continual RL Workshop (@continual_learn)
5
57
5,383
Turns out that many contemporary LLMs still struggle to explain probabilities and uncertainty in plain natural language 👀
1
11
2
14
3,467
The goal of this program is to identify under represented students from very early stage (high school) and introduce them to CS and programming! I’m super proud of this initiative that is getting mature over the last 3 years!
Chandar Lab (@ChandarLab) profile pictureChandar Lab@ChandarLabAug 6
For the third year in a row, we held our annual Highschool Internship for recent graduates from across Montreal! 🎓🎉💻 Ten bright students underwent a month-long summer school and got introduced & coached on the basics of computer science and programming!
Image from Chandar Lab (@ChandarLab)
6
23
3,882
For the third year in a row, we held our annual Highschool Internship for recent graduates from across Montreal! 🎓🎉💻 Ten bright students underwent a month-long summer school and got introduced & coached on the basics of computer science and programming!
Image from Chandar Lab (@ChandarLab)
1
8
2
25
5,263
@ChandarLab will be at #ICML 2026 in Seoul — South Korea 🇰🇷 !! Presenting our most recent state-of-the-art research on continual learning, reinforcement learning and AI agents! Drop by and check it out! 🔥
2
8
13
1,005
Excited to share our latest work: Loss Smoothing for Stable Adaptation Under Distribution Shift. TL;DR: When adapting a model, don’t abruptly switch to the new objective. Gradually smooth the transition from the old objective to the new one. Preprint: arxiv.org/abs/2607.00634 🧵 1/n
Loss Smoothing for Stable Adaptation Under Distribution Shift

Loss Smoothing for Stable Adaptation Under Distribution Shift

In settings such as fine-tuning and reinforcement learning, neural networks are often adapted under distribution shift. Standard adaptation methods typically optimize the target objective...

arxiv.org
1
15
3
50
12,687
We are super excited to announce @ChandarLab 7th Annual Research Symposium! 🔥 Join us on July 23–24 for two days of talks on deep learning, NLP, reinforcement learning, continual learning, and AI for science, plus a keynote by @mengyer from NYU! Event is hybrid!
2
14
2
12
2,763
📅 Submit by July 17 🦞
1
1
146
New paper from the lab: Ask a base LLM to pick a random weekday, and it might output "Wednesday" 80% of the time. We show that this can be fixed and that probabilistic calibration is actually a trainable capability.
1
10
4
23
5,970
Anabel retweeted
wta (@WTA) profile picture
wta
@WTAMay 17
A message from the Rome Champion! 📣 @ElinaSvitolina | #IBI26
6
23
2
431
10,553
Attention @arxiv authors: Our Code of Conduct states that by signing your name as an author of a paper, each author takes full responsibility for all its contents, irrespective of how the contents were generated. 1/
137
894
270
6,439
1,102,006
So happy to be a woman in AI, and more specifically, a woman at Mila! Shoutout to the scholarship committee for validating me before Reviewer 2 could invalidate me. Nice to know my 47 open tabs and sleep deprivation are contributing to science!
Mila - Institut québécois d'IA (@Mila_Quebec) profile pictureMila - Institut québécois d'IA
@Mila_QuebecMay 13
Mila is thrilled to unveil the 17 recipients of the third edition of its Equity, Diversity, and Inclusion (EDI) Scholarships Program. These students were chosen for their exceptional academic achievements, impactful research, and proactive dedication to fostering equity and inclusion both within the AI ecosystem and their local communities. A heartfelt congratulations to all! mila.quebec/en/news/mila-cel…
Image from Mila - Institut québécois d'IA (@Mila_Quebec)
Image from Mila - Institut québécois d'IA (@Mila_Quebec)
Image from Mila - Institut québécois d'IA (@Mila_Quebec)
2
5
608
Anabel retweeted
Nilaksh (@nilaksh404) profile picture
Nilaksh@nilaksh404May 12
Diffusion world models can help test and improve robot policies before running them on real robots. But can the choice of latent space make the WM more faithful? We show that semantic spaces beat reconstruction spaces on task relevant metrics. hskalin.github.io/semantic-w…
5
48
7
225
45,228
Load more