Cees Snoek@cgmsnoek

Oct 6

Pedro

@pmpcurvo

Oct 6

♾️ LiFT: Loop Flow Transformers 🧵👇

Diffusion models usually spend more inference compute by taking more sampling steps. What if we could also spend compute inside each step?

LiFT turns part of a DiT into a recurrent core, letting us scale depth at inference with only a nge. On ImageNet, it beats dense DiT-XL/2 with 3.34 lower FID, while using 60% fewer parameters, 32% fewer training FLOPs, and 52% fewer inference FLOPs.

Jointly led with Mohammad Mahdi Derakhshani 👑(@mmderakhshani)

And amazing supervisors Cees Snoek (@cgmsnoek), Jan-Willem van de Meent (@jwvdm) and Gertjan J. Burghouts (@gjburghouts)

Dima Damen@dimadamen

Oct 5

philip

@philiptorr

Oct 5

I care a lot about AI safety

But I do find the constant “AI will kill us all by the end of the decade” commentary increasingly unhelpful, and, frankly, a little tiresome.

Sometimes I feel certain people in Silicon Valley have lost touch with reality and normal people.

There are so many other pressing problems. A relatively small, wealthy, technologically immersed community in the Bay Area can spend an extraordinary amount of cognitive bandwidth on scenarios that feel overwhelmingly important from inside that community, while a normal person is worrying about housing, healthcare, war (especially with all the new wars that seem to have started **by humans** recently), crime, their children, energy prices, employment, climate, poverty, or simply getting through the week.

In fact, I think the more extreme rhetoric can be actively counterproductive. It collapses a very broad field: robustness, misuse, autonomy, cyber/bio risk, deception, control, systemic risk, military applications, concentration of power into one spectacular endpoint. And because the endpoint is so extreme and so uncertain, it encourages people either to become quasi-religiously preoccupied with it or to dismiss AI safety altogether.

And there is an important technical distinction between:

“The tail risk is serious enough that we should invest heavily in understanding and reducing it”

and

“The median outcome is human extinction within a few years.”

The first proposition can be justified even when the second is very poorly supported. Nuclear safety is an obvious analogy: you don't need to believe nuclear war is virtually certain next Tuesday to think command-and-control, escalation, verification and weapons safety deserve enormous attention.

I also think a mature AI-safety programme should deliberately cover several timescales. Immediate harms and misuse matter because they are happening now. Emerging agentic risks matter because systems are becoming more autonomous. And genuinely catastrophic or existential risks matter because even a relatively small probability multiplied by an enormous consequence warrants serious research. Those positions aren't in competition.

There is a further strategic problem with constant doom rhetoric: it makes calibration almost impossible. If every capability advance is “the last warning”, every model is “terrifying”, and every year is supposedly the crucial year before catastrophe, observers eventually discount the signal. Safety researchers ought to be unusually disciplined about uncertainty, evidence, falsifiability and updating—not less disciplined because the stakes are high.

(no AI was harmed in the making of this post)
repost if you agree to restore some sanity to the debate

Dima Damen@dimadamen

Sep 29

Silvia Zuffi@silvia_zuffi

Sep 29

A huge thanks to @AouadaDjamila for organizing @Euvip2026 and having me along with the amazing @dimadamen, @eliricci_ and @ftm_guney as kenote speakers!

Jitendra MALIK

@JitendraMalikCV

Sep 19

Toru

@ToruO_O

Sep 19

Inspired by recent thoughts from senior researchers whom I deeply respect (e.g. @JitendraMalikCV @Michael_J_Black @Ken_Goldberg @phillip_isola), I wrote down some thoughts as a junior researcher too.

The recent Astra demos in dexterous manipulation are truly impressive. But they s that I don’t think academia has good answers to yet:
- How should credit be assigned when an agent synthesizes many prior works into a new research result?
- How can academia attract and retain talents when so much of its incentive structure relies on credit?

I reflect on these questions in more detail in a blog post. Curious to hear what others think, and debates are very welcome!

toruowo.github.io/blog/posts/202…

Dima Damen@dimadamen

Sep 12

Dima Damen@dimadamen

Sep 12

It was a great week @eccvconf #ECCV2026 #Malmo for current members of the group and catching up with alumni! Here's a group pic missing a few,
@prajwal_gatti @Sid__Bansal @JacobChalkie @ahmad_darkhalil @zhifan_zhu @_alexstergiou @chiaraplizzari &Kevin, Toby, Evangelos.

Dima Damen@dimadamen

Sep 11

Dima Damen@dimadamen

Sep 11

This afternoon @eccvconf #ECCV2026 from 4pm
Find us at poster #363 presenting

Whareformer: Learning to Track What is Where in Long Egocentric Videos

Work w @naverlabseurope by @JacobChalkie @dlarlus @sinha_saptarshi and Yannis Kalantidis

Dima Damen@dimadamen

Sep 11

Jacob Chalk@JacobChalkie

Sep 10

Humans maintain a mental map of “what is where” for many objects, even if they are out of sight, but how does a model do this? That’s what we tackle in our #ECCV2026 paper, Whareformer, which we present tomorrow.
📍 Poster 363 · Session 4
🗓️ Friday 11 Sept 4:00–6:00 PM CEST
🧵1/3

Fabian Caba ∞@FabianCabaH

Sep 10

Dima Damen@dimadamen

Sep 10

Today @eccvconf #ECCV2026
Poster: #19 4:30PM (S2)
Gen2Balance: Generative Balancing for Long-Tailed Video Action Recognition
prajwalgatti.github.io/gen2balance/

Can we fill imbalanced action recognition datasets with generative videos to achieve SOTA? Yes we can
@prajwal_gatti @FabianCabaH