Anastasios Gerontopoulos

PhD Fellow, Archimedes Unit | University of Crete

[Anastasios Gerontopoulos @NasosGer]

1/n Multi-token prediction boosts LLMs (DeepSeek-V3), tackling key limitations of the next-token setup:
• Short-term focus
• Struggles with long-range decisions
• Weaker supervision

Prior methods add complexity (extra layers)
🔑 Our fix? Register tokens—elegant and powerful

[Anastasios Gerontopoulos @NasosGer]

@zmkzmkz Congrats!

[zed @zmkzmkz]

very excited to announce token order prediction has been accepted into ICML 2026!!

see you in Seoul!

very happy with how the paper turned out. we have an updated version of the paper on arxiv, and we’ll have a camera-ready version soon as well: https://t.co/cMOqnRjIFa

[Spyros Gidaris @SpyrosGidaris]

In ReDi, we showed that jointly modeling VAE latents + DINO features boosts diffusion.

CoReDi goes one step further: the feature space is no longer fixed — it’s learned.

→ Better joint modeling
→ 90× faster than SiT
→ 13× faster than REPA

https://t.co/Rg05UsbztF

[Thodoris Kouzelis @ThKouz]

1/n
Introducing CoReDi: Coevolving Representations for Joint Image–Feature Diffusion

Joint diffusion boosts image generation by injecting semantic features.
But one assumption goes unquestioned: the feature space is fixed.

What if it was learned instead? 🧵👇

[Spyros Gidaris @SpyrosGidaris]

Pixel or Latent World Models?

Why choose?

Re2Pix first predicts DINOv2 features for future frames, then conditions a diffusion model to render photorealistic frames.

→ more temporal semantic consistency & training efficiency (7–14× faster)

https://t.co/b2z5PE06PH

[Marawan Gamal @mrremila]

New preprint! Introducing ACTMat: Model Merging via Data-Free Covariance Estimation

Work done w/ @dtredsox13, @PTikeng, Colin Raffel and Guillaume Rabusseau

TL;DR: It's RegMean, but without needing data for covariance estimation (C ≈ Δᵀ Δ)

📄 https://t.co/qB0oafSxvq
🧵(1/6)

[Efstathios Karypidis @K_Sta8is]

1/n 🔀 Pixel or latent world models?

Video world models fall into two camps:
• generate photorealistic frames
• predict semantic features of the future (e.g., DINOv2)

Why choose one?

We introduce Re2Pix, a hierarchical approach that combines both. 🧵👇

[Greeks in AI @greeksinai]

🤖Thrilled to share the Call for Papers for Greeks in AI 2026

🇬🇷Greeks in AI 2026
📍Eugenides Foundation, Athens, Greece
📅15–17 July 2026

We invite submissions presenting recent advances, ongoing research, and emerging ideasi n Artificial Intelligence and Machine Learning

[Subham Sahoo @ssahoo_]

📢 New Paper: Scaling Beyond Masked Diffusion Language Models

Takeaways👇
🚫 Masked Diffusion ≠ endgame
🔥 Uniform-state beats AR on GSM8K
📉 Perplexity is a bad metric for comparing diffusion families

📰 https://t.co/H3AcTmkg79
🖥️ https://t.co/t3eKEdSvRH
🌐 https://t.co/RLCA9ivonL

collaborators: @jm_lemercier @zhihanyang_ @jdeschena @Jingyu227 @jwthickstun Ante Jukic

(1/5)

[Bill Psomas @bill_psomas]

1/n REGLUE Your Latents! 🚀

We introduce REGLUE: a unified framework that entangles VAE latents ➕ Global ➕ Local semantics for faster, higher-fidelity image generation.

Links (paper + code) at the end 👇

[Subham Sahoo @ssahoo_]

Thank you all for coming out to the discrete diffusion meetup. Turnout was over a 100 people😊

[Efstathios Karypidis @K_Sta8is]

I’m in San Diego attending NeurIPS 2025 this week! 🌴
Excited to present Dino-Foresight: Looking into the Future with Dino, our work on semantic future prediction. 🚀
Find us at Poster #4305
🕟 Tomorrow 4:30–7:30 PM PST
📍 Exhibit Hall C, D, E

[valeo.ai @valeoai]

Check out our works at @NeurIPSConf #NeurIPS2025 this week!
We present 5 full papers + 1 workshop about:
💡 self-supervised & representation learning
🖼️ generative image models
🧠 finetuning & understanding LLMs & multimodal LLMs
🔎 feature upsampling

https://t.co/9QIbm3ZttJ

[Anastasios Gerontopoulos @NasosGer]

Link to our original post on X:

[Anastasios Gerontopoulos @NasosGer]

1/n Multi-token prediction boosts LLMs (DeepSeek-V3), tackling key limitations of the next-token setup:
• Short-term focus
• Struggles with long-range decisions
• Weaker supervision

Prior methods add complexity (extra layers)
🔑 Our fix? Register tokens—elegant and powerful

[Anastasios Gerontopoulos @NasosGer]

Excited to share that I’ll be at NeurIPS 2025 in San Diego (Dec 2–6) presenting our poster paper on multi-token prediction! 🧠✨

📅 Poster presentation: Wednesday, Dec 3, 11 AM – 2 PM
📍 Location: Exhibit Hall C, D, E

[Anastasios Gerontopoulos @NasosGer]

If you’re interested in language models (autoregressive & diffusion), reasoning, or post-training, let’s connect!
DM me to grab a coffee or drop by our poster!
Looking forward to meeting new people. 🥳

#NeurIPS2025 #AI #LanguageModels #MachineLearning

[Divyat Mahajan @divyat09]

[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 embedding of the future sequence, not the tokens themselves

[Spyros Gidaris @SpyrosGidaris]

Three papers accepted to #NeurIPS2025 (one spotlight)! 🎉

Awesome works in generative modeling, multi-token prediction, and semantic future prediction.

Congratulations to all students and collaborators involved!
@NasosGer, @K_Sta8is, @ThKouz, @IoannisKakogeo1 & Nikos Komodakis!

[Anastasios Gerontopoulos @NasosGer]

@MinChonChiSF Thanks! I am glad that you liked our figure, it feels intuitive to us too.

[Anastasios Gerontopoulos @NasosGer]

Our paper on multi-token prediction with registers (MuToR) has been accepted to #NeurIPS2025 😀.
I am grateful to my co-authors, @SpyrosGidaris and Nikos Komodakis, for their guidance and collaboration. See you in San Diego!

If you are interested, check our detailed thread 👇

[Anastasios Gerontopoulos @NasosGer]

1/n Multi-token prediction boosts LLMs (DeepSeek-V3), tackling key limitations of the next-token setup:
• Short-term focus
• Struggles with long-range decisions
• Weaker supervision

Prior methods add complexity (extra layers)
🔑 Our fix? Register tokens—elegant and powerful

[Anastasios Gerontopoulos @NasosGer]

@Ella_Maru Thank you Ella!