
PhD with distinction in probabilistic deep learning @JagiellonskiUni. AI Research Scientist. Diffusion, AI4Science, GFlowNets, AI Safety. @Mila_Quebec alumni.
Joined January 2022
- Tweets276
- Following1.6K
- Followers316
- Likes59.7K
Pinned Tweet
1/ Two and a half weeks ago I defended my PhD at the @JagiellonskiUni in Kraków!
"Probabilistic deep learning: from efficient sampling to principled generation" - awarded with distinction. 🎓
Nine papers, three themes, one journey. Quick thread on what was inside 🧵
Marcin Sendera retweeted
Scientists at frontier AI labs have unique insight into the most advanced models, often seeing the associated risks months before models are released to the public. Their perspective is vital for keeping society informed and should be taken very seriously.
wsj.com/tech/ai/anthropic-re…
Marcin Sendera retweeted
In my latest op-ed for @TIME, I discuss why the OpenAI Hugging Face cyber incident represents a turning point for AI safety. If we want to prevent more autonomous cyberattacks from threatening critical infrastructure, we urgently need stronger regulatory oversight and new approaches to model training to ensure robust safety assurances by design, which is what we are tackling at @LawZero_.
Read the full piece: time.com/article/2026/09/09/…
Marcin Sendera retweeted
Replying to @crptAtlas
Super useful! I created a desktop app for similar purposes: Instead of blindly vibecoding, one can create a diagram determining i/o blocks with short descriptions. As a result, we get a structure for one's code. Check it here: github.com/jmtomczak/vibepro…
Marcin Sendera retweeted
New work led by @FlemmingKondrup and @tomjiralerspong highlights an important vulnerability in chain-of-thought monitoring for agents, I highly recommend giving it a read.
Link to the paper: arxiv.org/pdf/2603.16928
Can LLM agents figure out that their "private" chain of thought is being monitored, with no hints, no training, just from getting blocked?
Yes. And some then explicitly reason to hide their reasoning.
New paper with @FlemmingKondrup (co-first authors) & @Yoshua_Bengio: "Noticing the Watcher" 🧵
Marcin Sendera retweeted
Neural samplers for unnormalized densities have come a long way!
A year ago, sampling from a 2,500-D Gaussian mixture density with 20 random modes seemed to be completely out of reach without severe mode collapse.
Today, this is possible with Bridge Matching Samplers (an extension of stochastic interpolants to the data-free setting). Check out our poster at #ICML2026 to learn more! 👇
We present the Bridge Matching Sampler at ICML today (5pm, #3311), a version of stochastic interpolants that does not need data from the target, but only a density. We iterate reciprocal and Markovian projections and employ a fixed-point iteration, see openreview.net/pdf?id=gUbNUN….
Marcin Sendera retweeted
We present the Bridge Matching Sampler at ICML today (5pm, #3311), a version of stochastic interpolants that does not need data from the target, but only a density. We iterate reciprocal and Markovian projections and employ a fixed-point iteration, see openreview.net/pdf?id=gUbNUN….
Marcin Sendera retweeted
I just arrived in Seoul for #ICML, presenting three works on diffusion-based sampling:
Bridge Matching Sampler: openreview.net/pdf?id=gUbNUN…
Tensor Train Diffusion: openreview.net/pdf?id=DDQX97…
Discrete/continuous time equivalences: arxiv.org/pdf/2501.06148
Write me if you want to meet.
Marcin Sendera retweeted
For the last months, we’ve been working at @MoleculeOne with @OpenAI on something that felt like a glimpse of the future of science.
Together with @OpenAI , we achieved what we believe is the first near-autonomous discovery in organic chemistry.
openai.com/index/AI-chemist-…
🧵
Replying to @MarcinSendera @JagiellonskiUni @NeurIPSConf @TmlrOrg @UncertaintyInAI @ArnaudDoucet1 @AmosStorkey @PiotrRMilos @Yoshua_Bengio @Mila_Quebec @GMUMJU @MLinPL
11/
Currently at @JagiellonskiUni and NASK - National Research Institute on AI Safety: agentic red-teaming, reasoning models, probabilistic alignment.
1
5
151
12/
Always happy to chat about probabilistic methods, diffusion samplers, GFlowNets, AI for Science, or research more broadly!
🔗 Thesis & full bio: msendera.github.io/
Onward 🚀
3
7
176











