📢📢 I have a new paper in which I argue that the question “Can we make algorithms fair?” is a category error: https://t.co/fMDJUo2u6Q A few years ago I became disillusioned with algorithmic fairness research and stopped working in the area. So when I was invited to contribute

Introducing Bolmo, a new family of byte-level language models built by "byteifying" our open Olmo 3—and to our knowledge, the first fully open byte-level LM to match or surpass SOTA subword models across a wide range of tasks. 🧵

This is not just another strong open model. Nemotron actually releases training data (!), RL environments, and training code. This is a big difference: almost all model developers just want people to use their models; NVIDIA is enabling people to make their own models. We are

Just dropped a new text embedding methodology. Fast as heck on CPU only and still great for document similarity analysis, clustering, and classification. How? Use a tiny ReLU network to approximate a big transformer from lexical (term frequency / bag of words) features.

Since linear probes are popular again, maybe it’s a good time to point to the many issues with them, which were examined in detail in the NLP Interpretability community. The “mechanistic?” piece by @sarahwiegreffe and @nsaphra has many useful pointers. https://t.co/K7ehk5wMGr

Is this naivete or what people call engagement farming.

Unrelated to the prez receiving peace award from FIFA, my local Indian restaurant has given me this best NLP researcher award. So proud y’all 🤭

The US government shutdown of over 40 days has made a mess out of the NIH grant review system. The rescheduling need for 11K reviewers to discuss the applicaitons is a logistical nightmare. I am really grateful for the NIH SROs and other staff who are dealing with this 🙏

With recent big AI confs, good luck having facetime with folks who r any popular. Yes, they tweet they r open to meet at the conf, but most r too busy & may not be able prioritize unless u r also famous :-) This is from personal experience. Not blaming anyone, but this reality.

Starbucks going meta on my name 😀

👀Your small LMs (SLMs) are… not that fast? 🚀At NVIDIA Research, we release 𝐍𝐞𝐦𝐨𝐭𝐫𝐨𝐧-𝐅𝐥𝐚𝐬𝐡 (NeurIPS 2025), a hybrid SLM family designed around real-world latency and trained from scratch with 1B/3B sizes, achieving SOTA accuracy, latency, and throughput.

In biomedicine, even longer unusual abbreviations are overloaded. The following are two important commonly used terms that are abbreviated as EGFR. 1. estimated glomerular filtration rate 2. epidermal growth factor receptor

Vague posting or whatever it is called and I just saw someone doing a pre-preprint type thing where they tweeted out the abstract of their paper and some text without actually putting the preprint out and saying it will out in a few days. So tired of these weird antics 🙄

when you ask a PhD student when they’ll finish

We present Olmo 3, our next family of fully open, leading language models. This family of 7B and 32B models represents: 1. The best 32B base model. 2. The best 7B Western thinking & instruct models. 3. The first 32B (or larger) fully open reasoning model. This is a big

The highlight of “year in review” at AMIA this year (besides all the great papers) is this dystopian anime style video of AI taking over healthcare and locking humans out. It ended on a happy note though 😇 #amia2025

Got last minute invite for the Columbia #amia2025 mixer from @nychunhua (thank you!). It was a lot of fun chatting with her trainees and others students and faculty/alumni of @ColumbiaDBMI. Plus they had amazing vegan/vegetarian food 😀

Today we’re releasing Deep Research Tulu (DR Tulu)—the first fully open, end-to-end recipe for long-form deep research, plus an 8B agent you can use right away. Train agents that plan, search, synthesize, & cite across sources, making expert research more accessible. 🧭📚

Due to strong Chat<LLM> capabilities, this semester, all assessments (except final project) for my #nlproc course are in-person in-class, with access to SLP3 chapter & slide PDFs & any notes students take during lectures. Happy to share details with others (Please DM or email)

After 30 years of research and teaching, I found that I cannot teach any subject unless I have understood it fully. Hence I normally had to write a book before I teach any new course. Also, I found that, teaching is the best, probably only, way to do deeper and better research.