Daniel D'souza @ ACL ✈️🇦🇹
@mrdanieldsouza
Research Engineer @Cohere_Labs💙 | @UMichECE Alum 〽️ | 🇮🇳✖️🇺🇸 💫"The Universe Works in Mysterious Ways"💫
“Best Paper Award” @ ACL 2024 🪄What an incredible culmination of perseverance to connect and represent languages around the 🗺️! 🪄 🤗 Huge thanks to the @aclmeeting committee for recognizing the massive effort behind Project Aya @CohereForAI 💙 #ACL2024
I'm incredibly proud that Aya received #ACL2024 Best Paper Award 🥹. Huge congratulations to the Aya team and @CohereForAI community who make this possible by for extending frontiers of LLMs to multilingual, building Aya Model and Aya Dataset 🌿🌏
We have an incredible roster of accepted papers at @aclmeeting 2025. I will be there, as will many of our senior and engineering staff @mziizm @beyzaermis @mrdanieldsouza @singhshiviii 🔥 Looking forward to catching up with everyone.
There’s less than one week to go until @aclmeeting in Vienna, Austria! 🇦🇹 The Cohere Labs and @Cohere research teams are looking forward to showcasing some of our latest research and connecting with the community. Be sure to stop by our booth and say hello!
Me and the lovely folks at @Cohere_Labs 💙 & @cohere are at ACL all week! 🇦🇹 If you’ve been thinking about inference-time scaling, multilingualism or interested in staging a rebellion against “prompt engineering”👀, we should chat over a coffee! 😄☕️
This is really nice work, part of a wider effort to build intelligence that learns from context. arxiv.org/pdf/2506.14702
Prompt engineering places all the work on the end user to try and squeeze out performance. It's a hack to deal with the limitations in adaptability of our model. However, the future should be that this happens behind the scenes and is inferred automatically.
Prompts shouldn't have to be engineered. Our latest research marks another step towards fluid, natural language communication with LLMs.
Why wrestle with prompts for a specific task 🤹in a specific domain 🤹♂️ for a specific model family 🤹♀️? We propose a fundamental way we can teach models modes of interaction that are less brittle and a lot more adaptable at inference time!🪄⤵️
Prompts shouldn't have to be engineered. Our latest research marks another step towards fluid, natural language communication with LLMs.
Unimodal explainability tricks people into thinking a multimodal system uses every input. This paper builds three strict tests that force explanations to prove real cross‑modal reasoning. Today most explanation tools look at one data type at a time. That is fine for an…
Announcement: Papers in the park in Casablanca - Morocco is coming in the upcoming weeks Stay tuned 👀 @Cohere_Labs @sarahookr thank you for supporting 🫶🏻
Maybe shall we have one in Casablanca - Morocco too 🥹🥹🥹 cc : @Cohere_Labs are you open to help your community member as you always do?🥺
When Life Gives You Samples The Benefits of Scaling up Inference Compute for Multilingual LLMs
“When Life Gives You Samples: The Benefits of Scaling up Inference Compute for Multilingual LLMs” Led by: @ammar__khairi @mrdanieldsouza, @YeS855811, Julia Kreutzer , @sarahookr 📜Paper link: arxiv.org/abs/2506.20544
Can we improve the performance of LLMs during inference without the need for extensive sampling OR special reward models? 🤔 Our latest work introduces a new inference time scaling recipe that is sample-efficient, multilingual, and suitable for multi-task requirements. 🍋
🚨New Recipe just dropped! 🚨 "LLMonade 🍋" ➡️ squeeze max performance from your multilingual LLMs at inference time !👀🔥 🧑🍳@ammar__khairi shows you how to 1⃣ Harvest your Lemons 🍋🍋🍋🍋🍋 2⃣ Pick the Best One 🍋
🚀 Want better LLM performance without extra training or special reward models? Happy to share my work with @Cohere_labs : "When Life Gives You Samples: Benefits of Scaling Inference Compute for Multilingual LLMs" 👀How we squeeze more from less at inference 🍋, details in 🧵
💪🏼Huge thanks to my incredible mentors: Julia Kreutzer @mrdanieldsouza, @YeS855811, @sarahookr for guiding me and supporting this work ✨ Find our arXiv release here! 📜: arxiv.org/abs/2506.20544
We introduce two new selection techniques: CHOPs🥢 and X-MBR ⚖️, designed to amplify multilingual performance gains. Testing on 8B models (Aya-expanse, Qwen3), our methods achieve up to +12% win-rate vs greedy decoding on mArena-Hard from just 5 samples!
We're looking for a new member for the multilingual team with a focus on data engineering! Please apply at the link below:
The Multilingual Team at @cohere is hiring! If this sounds like you, please apply: - strong coding skills and a keen eye for detail - experience working with the challenges & joys of multilingual data Help us bring AI to the world! 🌏🌍🌎 jobs.ashbyhq.com/cohere/a87be94…
Yooo 👀 and there’s pizza!?! 🍕
Papers ✅ Park ⏳ Tomorrow at Trinity Bellwoods come learn about @Cohere_Labs new paper improving models performance on rare cases
We’re proud to have released 9 open models — all built to support research, experimentation, and real-world impact. 🌎 These models reflect our commitment to building powerful, accessible tools that can accelerate progress across machine learning and beyond.
🤝Arbitration is the future 🤝 “Why rely on a single teacher 🧑🏻🏫 when you can synthetically generate a much higher quality dataset by relying on specialized teacher models? 🧑🏻🏫👩🏫👨🏿🏫” Check out this fantastic summary of our recently accepted ACL 2025 work ✨
How can AI capture the nuances of different languages?💬🗨️ By using a team of specialized teacher models via Multilingual Arbitration we've achieved up to 19.5% improvement in win rates across languages. Find us at ACL to discuss how we can further break down language barriers.
🤹 How do we move away from complicated and brittle prompt engineering at inference for under-represented tasks?🤔 🧠 Our latest work finds that optimizing training protocols improves controllability and boosts performance on underrepresented use cases at inference time 📈