🔥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

We've had an account for Google Research for a while, but we're going to start posting more info about the work done by Google Research here. Follow for awesome research content!

SuperDial was the first company to deploy TensorZero in production. Really exciting to see the progress and impact they've made over the past year+! Congratulations to Sam Schwager, Harrison Caruthers, and the SuperDial team — well deserved! 🍾

Gemini 2.5 Pro is the most underrated model. Extraordinary intelligence for free. I am not sure why people don’t talk about this all the time.

Alternatively, pay DatologyAI to look at your data for you. This is what all our cracked scientists look like (from looking at the data).

NeurIPS workshop proposal rejected. We had AMAZING speakers, and a great list of organizers. No rebuttal phase. No feedback. No dense reward :( Since I work full-time on Gemini and have zero time to publish, this was my chance to contribute to academia 😞

It takes LITERALLY 2 min to setup gemini-cli and Im telling you its incredible. No need to setup credit card or anything. Its also super intutive and safe. you should definitely try it out 100% recommend ``` npm install -g @google/gemini-cli gemini ```

I converted one of my favorite talks I've given over the past year into a blog post. "On the Tradeoffs of SSMs and Transformers" (or: tokens are bullshit) In a few days, we'll release what I believe is the next major advance for architectures.

My favorite reading of the week by Yiding Jiang: Next era is not about learning from data but deciding what data to learn from. yidingjiang.github.io/blog/post/expl…

*Antidistillation Sampling* by Yash Savani Asher Trockman Zico Kolter et al. They modify the logits of a model with a penalty term that poisons potential distillation attempts (by estimating the downstream distillation loss). arxiv.org/abs/2504.13146

Tokenization has been the final barrier to truly end-to-end language models. We developed the H-Net: a hierarchical network that replaces tokenization with a dynamic chunking process directly inside the model, automatically discovering and operating over meaningful units of data

Tokenization is just a special case of "chunking" - building low-level data into high-level abstractions - which is in turn fundamental to intelligence. Our new architecture, which enables hierarchical *dynamic chunking*, is not only tokenizer-free, but simply scales better.

Big welcome to Varun Mohan and others from the Windsurf team joining Deepmind : )

0.030349% contribution achieved

prime intellect has acquired prime intellect

H-Nets are the future.

🎉 Excited to share that our paper "Pretrained Hybrids with MAD Skills" was accepted to Conference on Language Modeling 2025! We introduce Manticore - a framework for automatically creating hybrid LMs from pretrained models without training from scratch. 🧵[1/n]

Alex Robey Hamed Hassani Amin Karbasi Distillation 📉 Antidistillation 📈

For those at ICML, Audrey will be presenting this paper at the 4:30 poster session this afternoon! West Exhibition Hall B2-B3 W-1009

Huge congratulations to Vaishnavh, Chen and Charles on the outstanding paper award 🎉 We will be presenting our #ICML2025 work on creativity in the Oral 3A Reasoning session (West Exhibition Hall C) 10 - 11 am PT. Or please stop by our poster right after @ East Exhibition

1/ Today we announce Pleiades, a series of epigenetic foundation models (90M→7B params) trained on 1.9T tokens of human methylation & genomic data. Pleiades accurately models epigenetics for genomic track prediction, generation & neurodegenerative disease detection from cfDNA,