Angelo Gnazzo

PhD student @ TU Wien & AITHYRA | MSc Math @ETH | Interested in Graphs, Geometry, AI4Science, Generative AI

[Michael Bronstein @mmbronstein]

As Wotan would say: "vollendet das ewige Werk!"

https://t.co/ykUpGg7gWx

[Philipp Nazari @philna00]

Functional invariances are the basis of many transformer quantization methods.

Turns out there are way more for linear recurrences, like Mamba-2 and KDA!

Check out our new paper, LinQuant👇

[Michael Bronstein @mmbronstein]

Stellar lineup at the AITHYRA Next generation AI for Biomedical Science workshop -- next Monday 2 Nov at the IMBA lecture hall in Vienna Biocenter. No registration is required.

[Robert Pisarczyk @R_Pisarczyk]

Very excited to finally share what we've been building!

Today, we are launching Blatent Struct, our first language model. 🚀

One thing that's always bothered us about building with LLMs is how much engineering you need around the model. Getting an answer is one thing. Making sure it follows the right decision path, returns everything your application needs and doesn't break your business rules is another.

We wanted to change that.

Struct brings programmability to language models.

You can define complex, cascading decisions, conditional dependencies and exactly what each outcome must satisfy. You can also enforce business rules and permissions during inference.

And rather than stitching together multiple model calls, Struct can handle the whole decision tree in a single run.

We also spent a lot of time making sure this doesn't come at the expense of performance.

In our benchmarks, Struct achieves comparable accuracy to Claude Sonnet 5.5, at 11× the speed and over 100× lower estimated cost.

It's also over 2× faster than Jev, with higher average accuracy across 20 benchmarks. And on source-grounded extraction, it outperforms Cloudflare Clairic (87.5% vs. 81.7% across six benchmarks), at 62% lower estimated cost.

For me, the exciting part is what this enables: everything from document processing and customer support to agents that can make complex decisions while respecting the rules you've set.

This is just our first model, and there's a lot more to come.

I'd love for you to give it a try and tell us what you think. We're especially keen to hear what works, what doesn't and what you'd like us to build next.

🚀 Get started for free (up to 100m input tokens are on us) : https://t.co/iwEo6Cybo2

[Romeo Passaro @rom_pas_]

Happy to share our new work: Distributional Functional Rewards (DFR)! 🎉

We improve diversity and fidelity in diffusion fine-tuning 🔧✨

Big thanks to my coauthors @JeongjaeLee0, Jong Chul Ye, @mmbronstein, @Yoshua_Bengio, @AlexanderTong7 and @danyalrehman17! 🙏 Thread below 👇

[Jeongjae Lee @JeongjaeLee0]

🚨 New paper alert!
Despite recent advances in reward fine-tuning for diffusion and flow models, achieving high rewards remains challenging without sacrificing sample diversity or substantially drifting away from the pretrained distribution.
We address this with DFR: Distributional Functional Rewards, a framework that balances reward optimization with preserving desirable properties of the generated distribution! 🧵

[Zander Blasingame @zwblasingame]

Excited to announce the follow-up to our NeurIPS 2026 Oral: Improved Training of Strong Stochastic Flow Maps!

We improve the training of SSFMs by unlocking derivative-based losses!

Joint work with @antoniofrancaib @rmincam @ChristianKoke @Niklas_TR and @AlexanderTong7

[Rodrigo Casado Noguerales @rodcasnog]

🚨 New paper!

What if LLMs could reason like we do when solving Sudoku: keep a few candidates per cell and iteratively narrow them down?

Introducing Binary Discrete Diffusion for language and reasoning tasks: it denoises sets of candidate tokens using binary representations! 🧵

[Kieran Didi @DidiKieran]

Avengers assemble: 1 year BakerLab + NVIDIA GenAIR = 🔥 https://t.co/Cd2qiugCJo

Presenting RFD4-Proteina, our latest protein design model, significantly better and faster than everything we compared against, both in silico and in the lab! 1/30

[Federico Di Gennaro @fededg__]

Misalignment can emerge from recursive iterations of training AI on its own safe data, a phenomenon we call Progressive Misalignment.

We believe this failure raises important questions about how we should evaluate safety in AI systems which can evolve over time. 👇🏼🧵

[Karim Abdel Sadek @Karim_abdelll]

New surprising result:

We trained AI recursively on its own safe generations, while giving it a benign cue. Models became increasingly misaligned over time, giving advice on how to do harm and build biological weapons.

We call this Progressive Misalignment. 🧵👇

[Francesco Ruscio @francescoshox]

Very happy to share the first paper of my PhD: Normalizing Flow Maps! (NFM)⚡️

We make likelihood evaluation in flow maps fast and tractable by replacing the DiT backbone with a normalizing flow🔄.

A big thanks to @tim_hitge, @mmbronstein, @bose_joey and the rest of the team!

[Joey Bose @bose_joey]

🚨 New Paper Alert, 🎤 Accepted for an oral presentation at the NeurIPS 2026 STODY workshop!

📣 Introducing: Normalizing Flow Maps (NFM): Combining few-step transport from Flow maps with tractable likelihoods from Normalizing Flows.

This is my first work with my first crop of new PhD students, @francescoshox and @tim_hitge, who genuinely knocked this project out of the park.

📄 Paper: https://t.co/LCFBg1QjLT
📝 Blog: https://t.co/8dWEYwUEOm
💻 Code: https://t.co/Oroppp6NBR

🧵
1/8

[Niklas Rindtorff @Niklas_TR]

Introducing Langevin Flow Maps

TLDR: Neural Network Potentials that learn their own stochastic pathwise solution for faster simulation.

- Transferable across molecular systems.
- Transition dynamics are preserved.
- No trajectory data are required.

Work together with @sgmccallum, supervised by @AlexanderTong7, and @JamesFosterBath.

Arxiv: https://t.co/GGgujg56YH
Data: https://t.co/TeYvKN13o7
Code: https://t.co/HYiMqTI4jT

[Rodrigo Casado Noguerales @rodcasnog]

Super excited to present "What Does a Discrete Diffusion Model Learn?" as an oral at the DiffuLM (Diffusion Language Models) workshop at NeurIPS 2026 in Sydney! If you'll be there and want to chat about discrete diffusion, feel free to reach out!

[Luca Ambrogioni @LucaAmb]

Bitstream language diffusion is cool, but what about bistream diffusion for everything?

Happy to share:
Multimodal Co-Generation With Continuous Bitstream Diffusion

Highly reliable and high performance training on many modalities without any modality-specific calibration!

[İsmail İlkan Ceylan @ismaililkanc]

One model, wildly different graphs and tasks, and some very intriguing generalization capabilities.

Excited about the future of this! Fantastic work by @louistichelman with great contributions from the usual suspects @hxyscott and @jw9730!

[Louis @louistichelman]

Can one foundation model reason across different graphs, features, relation types and prediction tasks?

Introducing 𝗪𝗮𝗻𝗱𝗲𝗿🚶

Our new preprint explores 𝗴𝗿𝗮𝗽𝗵 𝗳𝗼𝘂𝗻𝗱𝗮𝘁𝗶𝗼𝗻 𝗺𝗼𝗱𝗲𝗹𝘀 through a shared prediction objective and a random-walk interface.

🧵 1/7

[Federico Di Gennaro @fededg__]

🚀 Science is the next big challenge for agentic AI systems.

While LLMs are becoming incredibly capable, it is still an open question how we can translate their intelligence into scientific discoveries, where closing the loop is costly and reliable reward functions are often unavailable.

We introduce LLM-to-Flow Distillation (LFD): a method that combines the scientific knowledge of frontier LLMs with the structural priors of pretrained generators, allowing us to steer these latter toward natural-language discovery goals.

Check out our work in the thread 👇🏼

[Angelo Gnazzo @gna_ngelo]

Our new preprint LLM-to-Flow Distillation (LFD) is finally live! 🧬

I’m convinced language is the right interface for controlling generative models. And that should apply to molecules just as much as images!

LLMs know a lot about science. DPO gives us a way to transfer that knowledge into scientific generators. With LFD, “design molecules that bind to the D2 dopamine receptor” becomes a goal we can actually train a flow model to pursue.

The LLM builds an adaptive curriculum around what the generator can currently produce. As the model improves, the teaching and guidance from the LLM change too.

The cool part is that we can do this with models we already have, without redesigning their architectures or engineering a task-specific proxy reward.

Really happy to share this work with Riccardo, Federico and the rest of the team!! Full thread below 👇

[Riccardo De Santi @r_de_santi]

We’re bringing natural-language control to any scientific generator.

We are very excited to introduce LLM-to-Flow Distillation (LFD), a method to adapt scientific generators to high-level natural-language discovery goals.

Starting from textual goals like “design peptides with promising anticancer activity” or “design small molecules binding to the D2 dopamine receptor”, LFD distills high-level goal-related knowledge from LLMs into scientific generators through iterative DPO on LLM-generated sub-goals.

🔥 LFD does not need task-specific proxy rewards, works with any frontier LLM via API-access, and outperforms state-of-the-art reward-guided methods such as FlowGRPO and DiffusionNFT on held-out computational evaluations. (1/6)