David Ifeoluwa Adelani 🇳🇬

Hallelujah! I’m excited to share that I’ve been selected as a 2025 AI2050 Early Career Fellow by @Schmidtsciences This year’s fellows represent 42 institutions across eight countries, working to ensure AI benefits humankind. Learn more at: lnkd.in/eZA5FHci

Fresh on arXiv! 😁 Our new paper reformulates tokenisation as a linear program (LP), which we solve to get SOTA tokenisers! As a bonus, this LP allows us to know how close to optimal any tokeniser is! Check it out! 👇

Some new results I found surprising that I’m tweeting for Chris (who isnt on here). With enough compute, the best data filter for LMs (on DCLM) might be no filter. Why? Large models can tolerate a surprising amount of nominally 'low quality' data, and can sometimes even benefit.

🔥The VoiceMOS Challenge 2026 kicks off today! 🔥 Please register using the link below: forms.gle/L6YdkUf1PJdSSw… We will send you the challenge information afterwards! Friday, July 31: Evaluation dataset release. Friday, August 7: Predicted scores submission deadline.

🤖 AI text detectors are widely deployed in education and integrity workflows, but what are they actually tracking? We report a surprising finding: text from base models is overwhelmingly judged as human by GPTZero and Pangram. 👇 (1/6)

#NLProc Slides for our LREC tutorial are now available online: Multilingual and Multimodal LLMs in the Wild: Building for Low-Resource Languages Slides: mm-llms-in-the-wild.github.io/content/ Reading list: mm-llms-in-the-wild.github.io/reading_list/ Includes resources on multilingual, multimodal, and low-resource language technologies. W/ @shammur_absar Enamul Haque #LREC2026 #NLProc #MultimodalAI #LLMs

🌍 Introducing #GenAI4World Workshop Generative AI for the World: The First Workshop on Globalizing Tasks, Evaluations, and Systems at #COLM2026 📅 October 9, 2026 📍 San Francisco, Co-located with @COLM_conf 🔗 sites.google.com/view/genai4wor… Stay tuned for the updates!

🚨 𝗖𝗮𝗹𝗹 𝗳𝗼𝗿 𝗣𝗮𝗽𝗲𝗿𝘀 – #GenAI4World Workshop @COLM_conf #COLM2026 We invite papers on multilingual, multimodal, and culturally grounded LM 🛠️ Two non-archival tracks: 1️⃣ Workshop Track 2️⃣ Conference Track 🔗 Submit: openreview.net/group?id=colmw… Details below ⬇️

Attention @arxiv authors: Our Code of Conduct states that by signing your name as an author of a paper, each author takes full responsibility for all its contents, irrespective of how the contents were generated. 1/

The penalty is a 1-year ban from arXiv followed by the requirement that subsequent arXiv submissions must first be accepted at a reputable peer-reviewed venue. 4/

Nice work, good to see some improvements on AfriMGSM

Introducing AutoScientist. Most model training fails outside of frontier labs. AutoScientist automates the full research loop so it doesn't have to.

We are looking for a postdoctoral researcher in speech and audio processing, with a possible start in the Fall 2026 semester. If you are interested in working with us, please apply through the following form: forms.gle/gfENMMrRf1nmnT…

Today we're sharing our work on interaction models. A new class of model trained from scratch to handle real-time interaction natively, instead of gluing it onto a turn-based one. youtu.be/A12AVongNN4

Fast Byte Latent Transformer is accepted to ICML 2026! ⚡🥪 Byte-level LMs promise to free us from subword tokenizers, but decoding one byte at a time is super slow. We make BLT generation more efficient with BLT-D: text diffusion for parallel byte decoding. 1/

🌍✨ We’re thrilled to welcome Dr David Adelani @davlanade, Assistant Professor at the @mcgillu McGill University School of Computer Science and Canada CIFAR AI Chair @CIFAR_News, as our invited speaker at this year’s Deep Learning Indaba in Lagos, Nigeria! Dr Adelani will be delivering a talk on Monday, August 3rd, at 12:00 pm GMT+1 titled: Lessons from the Trenches: Building African-Centric LLMs for African Languages. An award-winning researcher with over 50 publications, Dr Adelani has led the development of groundbreaking models like AfroXLMR and AfriqueLLM. His work is at the forefront of ensuring that low-resource African languages are prioritised in the global advancement of AI technologies. Reflecting on the vision for his talk and the spirit of the Indaba community, Dr Adelani shares: "Ultimately, this talk aims to inspire more African researchers to develop African-centric models tailored to their languages and real-world use cases." Join us for this essential conversation on the future of African AI! Registrations for the Virtual Indaba are still open! deeplearningindaba.com/2026/virtual-i… #DLI2026 #Indaba2026

Can you boost your AI review scores by asking an LLM to rewrite your paper? Yes! We call it paper laundering Our @icmlconf spotlight paper argues current AI reviewers aren't ready to automate peer review, and outlines what a science of peer review automation should look like🧵👇

Deadline extension! - The task is simple: get audio + its translation and estimate how good it is. - Mark your name as the winner of the first Speech Translation Metrics Shared Task at IWSLT 2026 🏆 Predictions submission: May 7, 2026 Description paper: May 10, 2026

Will be presenting at #LREC2026 🌍 Our half-day tutorial — "Multilingual and Multimodal LLMs in the Wild: Building for Low-Resource Languages" — will walk through how to build LLM systems that see, hear, and read across the languages and dialects that rarely get first-class support. What we'll cover (14:00–18:00 CET): • The model landscape — recent VLLMs, AudioPaLM and Omni-models • Low-cost data creation — translation/back-translation, OCR/ASR bootstraps, safety and culture-aware filtering • Efficient training — adapters, LoRA/QLoRA, quantization, and MoE routing for modality/language specialization • Practical speech→text→LLM pipelines — cascaded vs. unified speech–text systems • Culture-aware evaluation — xGQA, MaRVL, HaVQA, dialectal stress tests, hallucination and grounding diagnostics • Hands-on recipes, slides, and lab notebooks to take home Presented by: 🎙 Firoj Alam — Qatar Computing Research Institute, HBKU 🎙 Shammur Absar Chowdhury — Qatar Computing Research Institute, HBKU 🎙 Enamul Hoque Prince — York University 📅 12 May 2026 · 14:00–18:00 CET 🔗 mm-llms-in-the-wild.github.io If you work on multilingual NLP, speech, vision-language models, or inclusive AI — please drop by. We'd love to learn with you. #LREC2026 #MultimodalAI #MultilingualNLP #LLM #LowResourceLanguages #SpeechAI #VisionLanguage #NLProc #ResponsibleAI W/ @shammur_absar Enamul Haque

Here is more precise submission policy. For full details, please check the latest CFP on our website! 2026.emnlp.org/calls/main_con…

NEW paper from Apple. Interesting idea: "Attention to Mamba". The paper introduces a two-stage recipe for cross-architecture distillation from Transformers into Mamba. Naive distillation collapses teacher performance. Their trick: first distill the transformer into a linearized-attention student using a kernel adaptation, then transfer that student into a pure Mamba with no attention blocks. On a 1B model trained on 10B tokens, the Mamba student hits 14.11 perplexity against a 13.86 Pythia-1B teacher, nearly matching quality at linear-time inference cost. If you can reliably convert trained transformers into state-space models without retraining from scratch, the entire open-weights ecosystem becomes cheaper to serve at long context. This is the kind of quiet infrastructure work that decides which architectures actually get deployed in agent stacks. Paper: arxiv.org/abs/2604.14191 Learn to build effective AI agents in our academy: academy.dair.ai