Arkil Patel
@arkil_patel
Followers
1K
Following
2K
Media
20
Statuses
226
CS PhD Student at Mila and McGill | Worked at AllenNLP and Microsoft Research
Montrรฉal, Quรฉbec
Joined October 2016
๐๐ก๐จ๐ฎ๐ ๐ก๐ญ๐จ๐ฅ๐จ๐ ๐ฒ paper is out! ๐ฅ๐ We study the reasoning chains of DeepSeek-R1 across a variety of tasks and settings and find several surprising and interesting phenomena! Incredible effort by the entire team! ๐: https://t.co/CDlFHD28xQ
Models like DeepSeek-R1 ๐ mark a fundamental shift in how LLMs approach complex problems. In our preprint on R1 Thoughtology, we study R1โs reasoning chains across a variety of tasks; investigating its capabilities, limitations, and behaviour. ๐: https://t.co/Cyy18kYQ45
1
5
26
๐จHow do LLMs acquire human values?๐ค We often point to preference optimization. However, in our new work, we trace how and when model values shift during post-training and uncover surprising dynamics. We ask: How do data, algorithms, and their interaction shape model values?๐งต
2
49
124
[1/9] While pretraining data might be hitting a wall, novel methods for modeling it are just getting started! We introduce future summary prediction (FSP), where the model predicts future sequence embeddings to reduce teacher forcing & shortcut learning. ๐Predict a learned
10
47
221
Excited to share our new work on the expressivity of Transformer-based multi-agent systems and understanding the trade-offs in communication, no. of agents, and achievable speedups โจ Work led by @frisbeemortel; check out his thread for details!
Is there such a thing as too many agents in multi-agent systems? It depends! ๐งต Our work reveals 3 distinct regimes where communication patterns differ dramatically. More on our findings below ๐ (1/7)
0
4
13
Itโs clear next-gen reasoning LLMs will run for millions of tokens. RL at 1M needs ~100ร compute than 128K. Our Markovian Thinking keeps compute scaling linear instead. Check out Miladโs thread; some of my perspectives below:
Introducing linear scaling of reasoning: ๐๐ก๐ ๐๐๐ซ๐ค๐จ๐ฏ๐ข๐๐ง ๐๐ก๐ข๐ง๐ค๐๐ซ Reformulate RL so thinking scales ๐(๐ง) ๐๐จ๐ฆ๐ฉ๐ฎ๐ญ๐, not O(n^2), with O(1) ๐ฆ๐๐ฆ๐จ๐ซ๐ฒ, architecture-agnostic. Train R1-1.5B into a markovian thinker with 96K thought budget, ~2X accuracy ๐งต
19
93
896
Introducing linear scaling of reasoning: ๐๐ก๐ ๐๐๐ซ๐ค๐จ๐ฏ๐ข๐๐ง ๐๐ก๐ข๐ง๐ค๐๐ซ Reformulate RL so thinking scales ๐(๐ง) ๐๐จ๐ฆ๐ฉ๐ฎ๐ญ๐, not O(n^2), with O(1) ๐ฆ๐๐ฆ๐จ๐ซ๐ฒ, architecture-agnostic. Train R1-1.5B into a markovian thinker with 96K thought budget, ~2X accuracy ๐งต
14
202
919
Iโm at CoLM this week! Come check out our work on evaluating RMs for agent trajectories! These days, Iโm thinking about forecasting generalization, scaling laws, and safety/adversarial attacks. Ping me if you wanna chat about research!
i will be presenting AgentRewardBench at #COLM2025 next week! session: #3 date: wednesday 11am to 1pm poster: #545 come learn more about the paper, my recent works or just chat about anything (montreal, mila, etc.) here's a teaser of my poster :)
0
5
7
Check out this new work on techniques for constructing Transformers for algorithmic tasks! Excited to have been part of this project!
We present The Transformer Cookbook: a collection of recipes for programming algorithms directly into transformers! Hungry for an induction head? Craving a Dyck language recognizer? We show you step-by-step how to cook up transformers for these algorithms and many more!
0
3
8
Here's a list of recommendations for what to do in Montreal during @COLM_conf and beyond. ๐: https://t.co/9ixyy2Y7Yl Many thanks to my co-authors @benno_krojer and @frisbeemortel.
github.com
A list of things to do in Montrรฉal. Contribute to mmarius/montreal-things-to-do development by creating an account on GitHub.
Who will be at @COLM_conf ? I'm preparing a list of recommendations for what to do in beautiful Montreal. Stay tuned. ๐ฅฏ
6
18
64
Exciting news! We're thrilled to announce the appointment of Professor @hugo_larochelle as Mila's new Scientific Director! A deep learning pioneer and former head of Google's AI lab in Montreal, Hugo's leadership will be pivotal in advancing AI for the benefit of all. Read the
12
29
257
Come by our #ACL2025 poster tomorrow to discuss the safety risks surrounding increasingly capable instruction-following retrievers (or anything safety related)! 16:00-17:30 on Tuesday in Hall 4/5
Come and visit our poster on the Safety of Retrievers @aclmeeting ๐๏ธTuesday, Findings Posters, 16:00-17:30 ๐จInstruction-following retrievers will become increasingly good tools for searching for harmful or sensitive information.๐จ
0
4
16
Come and visit our poster on the Safety of Retrievers @aclmeeting ๐๏ธTuesday, Findings Posters, 16:00-17:30 ๐จInstruction-following retrievers will become increasingly good tools for searching for harmful or sensitive information.๐จ
Instruction-following retrievers can efficiently and accurately search for harmful and sensitive information on the internet! ๐๐ฃ Retrievers need to be aligned too! ๐จ๐จ๐จ Work done with the wonderful @ncmeade and @sivareddyg ๐ https://t.co/yLJPiy1d0j Thread: ๐งต๐
1
7
19
@aryopg Nice work! We observed a similar trend on certain math tasks in our work: https://t.co/hNlFcjKauc Section 4.1 has a discussion of our findings. You might want to consider citing it :) cc @saraveramarjano @arkil_patel @sivareddyg
0
5
13
If youโre at ICML and if you work on interpretability or causality, go talk to @_shruti_joshi_, she has a fantastic paper!
I will be at the Actionable Interpretability Workshop (@ActInterp, #ICML) presenting *SSAEs* in the East Ballroom A from 1-2pm. Drop by (or send a DM) to chat about (actionable) interpretability, (actionable) identifiability, and everything in between!
0
0
3
SafeArena is being presented at #ICML2025 !! Check out our poster and talk to @ncmeade for all things โsafety โช agents โช LLMsโ!
I'll be at #ICML2025 this week presenting SafeArena (Wednesday 11AM - 1:30PM in East Exhibition Hall E-701). Come by to chat with me about web agent safety (or anything else safety-related)!
0
1
11
Congrats @vernadankers!! Weโre lucky to have you join our lab!
Congratulations Verna! This was one of the best theses I've ever read, I highly recommend checking out Verna's work on the tradeoffs between memorization and generalization in language models!
0
0
5
I miss Edinburgh and its wonderful people already!! Thanks to @tallinzen and @PontiEdoardo for inspiring discussions during the viva! I'm now exchanging Arthur's Seat for Mont Royal to join @sivareddyg's wonderful lab @Mila_Quebec ๐คฉ
Huge congratulations to Dr. @vernadankers for passing her viva today! ๐ฅณ๐ It's been an honour sharing the PhD journey with you. I wasnโt ready for the void your sudden departure left (in the office and in my life!). Your new colleagues are lucky to have you! ๐ฅบ๐ฅฐ @Edin_CDT_NLP
11
11
100
"Build the web for agents, not agents for the web" This position paper argues that rather than forcing web agents to adapt to UIs designed for humans, we should develop a new interface optimized for web agents, which we call Agentic Web Interface (AWI).
9
59
197
Do LLMs hallucinate randomly? Not quite. Our #ACL2025 (Main) paper shows that hallucinations under irrelevant contexts follow a systematic failure mode โ revealing how LLMs generalize using abstract classes + context cues, albeit unreliably. ๐ Paper: https://t.co/YEK4TaI7pq 1/n
6
26
44
๐ข New Paper! Tired ๐ด of reasoning benchmarks full of math & code? In our work we consider the problem of reasoning for plot holes in stories -- inconsistencies in a storyline that break the internal logic or rules of a storyโs world ๐ W/ @melaniesclar, and @tsvetshop 1/n
3
54
262