Benno Krojer
@benno_krojer
Followers
2K
Following
56K
Media
352
Statuses
6K
AI phding @Mila_Quebec @mcgillu (past: @AIatMeta). Interests: interpretability, language grounding (V+L), evals, reasoning. Vanier Scholar. ⚽🥨🧩
Montréal, Québec
Joined June 2014
Excited to share the results of my internship research with @AIatMeta, as part of a larger world modeling release! What subtle shortcuts are VideoLLMs taking on spatio-temporal questions? And how can we instead curate shortcut-robust examples at a large-scale? Details 👇🔬
Our vision is for AI that uses world models to adapt in new and dynamic environments and efficiently learn new skills. We’re sharing V-JEPA 2, a new world model with state-of-the-art performance in visual understanding and prediction. V-JEPA 2 is a 1.2 billion-parameter model,
3
24
64
Honored to receive the Computer Science Canada Outstanding Early Career Researcher award 🏅. It is a recognition of the work carried out by my students for their courage to push fundamental ideas in natural language processing even in the era of LLMs. Thanks to my mentors and
Congratulations to Siva Reddy (@sivareddyg), Core Academic Member at Mila, who has received the prestigious Outstanding Early Career Computer Science Researcher Award from @CSCan_InfoCan , the leading organization for the computer science community in Canada.
9
8
98
One of the most interesting (and hardest) questions a researcher can ask: Why has nobody tried this earlier? It is often hard to understand what exactly were the technical roadblocks or intellectual lack of thinking outside the box. In retrospect it always seems obvious. I am
0
1
14
Come to Montreal if you wanna work in cozy cafes while the world outside is snowy and shiny Working from Mila is also fun but today I needed to visit my favorite cafe I hadn't been to in weeks
Mila's annual supervision request process is now open to receive MSc and PhD applications for Fall 2026 admission! For more information, visit https://t.co/r01eLcY1P4
3
1
23
Can't wait to dive into this paper! Their 4 stages of cognition made me think back to one of my early favorite papers (Experience Grounds Language) and their 5 World Scopes. Not too related in the details, but both give the same vibe of a paper you want to read deeply
Introducing Cambrian-S it’s a position, a dataset, a benchmark, and a model but above all, it represents our first steps toward exploring spatial supersensing in video. 🧶
0
0
9
My lab mate @bhatia_mehar recently finished a cool project: Compared to humans, with AI models we have the unique chance to go back in time and change specific aspects during training --> how does that affect values/alignment?
🚨How do LLMs acquire human values?🤔 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?🧵
1
2
10
Sometimes I have fun outside of work: My friend Jack and I do a "Sports Day" roughly once a year, it's a ton of fun So this year we did 26 sports in one day (1 for each letter A to Z) and Jack did an amazing job editing it all together: https://t.co/cGRcR2Cd3f This day will
0
0
5
how did i ever have time-intensive hobbies outside of my phd? hard to imagine how i managed it just 1-2 years ago
0
0
3
🚨 Interested in generative world models? We’re thrilled to host Stephen Spenser (@GoogleDeepMind) at the World Modeling Workshop 2026, where he’ll talk about the Genie series of models! 🌐 https://t.co/inI2YV2Dsl
3
17
115
Can't stand these engagement bait tactics, I really think the credit here should go to the folks at @BasisOrg (led by @ZennaTavares) and not the one MIT co-author who happens to be involved.
🔥 MIT just exposed every top AI model and it’s not pretty. They built a new test called WorldTest to see if AI actually understands the world… and the results are brutal. It doesn’t just check how well a model predicts the next frame or maximizes reward it tests whether it
4
6
142
[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. 📌Predict a learned
10
46
217
Revisiting last week’s open problems scandal, I wrote about LLMs as Lore Laundering Machines and why some are blind to the novelty whitewashing.
argmin.net
When insight comes from willful forgetting.
3
16
74
Our fully open Olmo models enable rigorous, reproducible science—from unlearning to clinical NLP, math learning, & fresher knowledge. Here’s how the research community has leveraged Olmo to make the entire AI ecosystem better + more transparent for all. 🧵
2
4
42
Great take-away from @jeffclune's tea talk at @Mila_Quebec: Open-ended curiosity-driven research is worthwhile and very important
0
4
28
In which research areas do people still use continuous tokens/soft prompts? For example in vision we map vision tokens into LLM prefix space Who still does soft prompt tuning? Or adversarial attacks with soft prompt vectors? Or maybe other modalities than vision (audio, bio
0
1
4
Great open-source news for VL folks
🚨 New paper out! “FineVision: Open Data Is All You Need” 🥳 We unified 200+ data sources into 24M samples. That’s 17.3M images and 9.5B answer tokens, the largest open VLM dataset ever released. All fully documented, reproducible, and available for everyone. And there's more! 🎢
0
0
7
My computer science college professors used to look down on programming -- it's too volatile to deserve formal education (new paradigms and languages come out all the time). Instead, we did all kinds of theory: types, probability, queuing. Back then, it felt anachronistic and
Harvard and Stanford students tell me their professors don't understand AI and the courses are outdated. If elite schools can't keep up, the credential arms race is over. Self-learning is the only way now.
262
1K
13K
i recently fell in love with interpretability as a field and Naomi is one of the top people in the world you could work with and learn from!
I’m recruiting PhD students for 2026! If you are interested in robustness, training dynamics, interpretability for scientific understanding, or the science of LLM analysis you should apply. BU is building a huge LLM analysis/interp group and you’ll be joining at the ground floor.
0
0
11
Couldn't have wished for a better place to do my PhD, come apply!
Mila's annual supervision request process is now open to receive MSc and PhD applications for Fall 2026 admission! For more information, visit https://t.co/r01eLcY1P4
0
0
9
Fun fact: mics for podcasting are way harder than we thought. Sorry for the audio! We are still working on it I suppose it's like in ML research: simple baselines are hard to beat... we have not been able to get better than the built-in macbook microphone
0
0
4