🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

🚨Preprint Alert As promised, announcing the Aya Vision Technical Report – detailing the recipe to build SOTA multilingual multimodal models.

How do we build multimodal systems that work effectively across the globe? 🌍 Today we release the Aya Vision Technical Report, the detailed recipe behind Aya Vision models, unifying state-of-the-art multilingual capabilities in multimodal and text tasks across 23 languages!

‼️ 1/n Ask your reasoning model to think in lower resource language does degrade models’ performance at the moment. My awesome Co-author already communicated the main points in the thread, I will just communicate some random things we learned in my 🧵

🧵 Multilingual safety training/eval is now standard practice, but a critical question remains: Is multilingual safety actually solved? Our new survey with Cohere Labs answers this and dives deep into: - Language gap in safety research - Future priority areas Thread 👇

We're happy to announce that effective as of July 1, 2025, faculty members Stephen Bach and Srinath Sridhar have received named chairs. Steve is now the Eliot Horowitz Assistant Professor in CS and Srinath is the John E. Savage Assistant Professor in CS: cs.brown.edu/news/2025/06/0…

Excited to release Trove, our lightweight, hackable, and easy-to-use toolkit for dense retrieval, which simplifies experiments. Trove makes it easy and efficient to explore all the combinations of your documents and queries for training, and streamlines multi-node evaluations.

Excited to share more about Spurious Rewards! Also keep an eye out for some new experiments and arxiv coming soon 👀🔜

these are great takes! really cleared things up for me

How can we make language models more flexible to adapt to new languages after pretraining? 🌏 🧠 Our latest work investigates whether a tokenizer trained on more languages than the pretraining target can improve language plasticity without compromising pretraining performance.

🚨New pretraining paper on multilingual tokenizers 🚨 Super excited to share my work with Cohere Labs: One Tokenizer To Rule Them All: Emergent Language Plasticity via Multilingual Tokenizers

Huge congrats to Diana Abagyan on her first first author paper. Was a pleasure collaborating on this work — we ask what cheap interventions in pre-training can allow for more language plasticity downstream.

Thanks AK for the spotlight on our work I really believe strongly in this wider direction — of taking the pressure off everyday users to be master prompt engineers and inferring controllability directly from tasks.

Can we train models for better inference-time control instead of over-complex prompt engineering❓ Turns out the key is in the data — adding fine-grained markers boosts performance and enables flexible control at inference🎁 Huge congrats to Daniel D'souza  for this great work

Amir Efrati Stephanie Palazzolo Worth reading this research which showed it has already been turned into cheat-slop and that meta was one of the worst culprits for gaming it x.com/singhshiviii/s…

We see so many work this week about "emergent misalignment", but how is it fundamentally different from LLM jailbreaking research? I wrote a short blog post about it: yongzx.substack.com/p/emergent-mis…

Last week, I gave a talk at Tsinghua University on scaling test-time compute with generative process verifiers/PRMs that verify reasoning with reasoning. jietang thank you for the invite! TL;DR: A super data-efficient recipe to train generative process verifiers that can scale

Proud to support South Korean research. 🇰🇷 Our new grant program is designed to encourage and support the Korean research ecosystem and social impact initiatives. Let’s innovate together.