The core tech of the current AI breakthroughs is as old as AI itself: the perceptron was already invented in 1957! How can we improve on this? Liquid.ai is a new MIT startup that rethinks function approximation, using Liquid Neural Networks: linkedin.com/posts/joschaba…

Hello World! Excited to be out of stealth mode! Let me introduce you to our Liquid team: Joscha Bach Joscha Bach (MIT, Harvard, Intel) Jimmy Smith Jimmy Smith (Stanford) Stefano Massaroli Stefano Massaroli (MILA) Paul Pak Paul Pak (NASA, UW-Madison) Noel Loo

☀️Thrilled to unveil our latest work to appear at #NeurIPS2023, on improving solar irradiance forecasting! We present CrossViViT, an architecture that uses cross-attention to combine satellite data and time-series modalities for improved accuracy (arxiv.org/abs/2306.01112). 1/10🧵

Announcing StripedHyena 7B — an open source model using an architecture that goes beyond Transformers achieving faster performance and longer context. It builds on the lessons learned in past year designing efficient sequence modeling architectures. together.ai/blog/stripedhy…

We've been hard at work pushing the frontiers of efficient architecture design and optimization. StripedHyena-7B is the result: the first alternative architecture truly competitive with the best Transformers of its size or larger. And it's very fast.

TogetherAI just released StripedHyena 7B. At first look ~Mistral level model with significant memory and speed benefits: together.ai/blog/stripedhy… If you are interested in how it works, I surveyed this literature earlier this year in an MLSys keynote: youtube.com/watch?v=dKJEpO…

Excited to announce StripedHyena!🚀 This is the pinnacle of a long line of efficient architecture research to which I’ve been honored to contribute. Kudos to my dear friend Michael Poli for bringing our crazy ideas to the world, and Together AI for fostering our research

My core ML team (AI at Meta) is hiring research interns! Our projects span optimization, optimal transport, optimal control, generative modeling, complex systems, and geometry. Please apply here and reach out ([email protected]) if you're interested: metacareers.com/jobs/627997209…

HyenaDNA poster went awesome! Honestly had sooooo much fun sharing the work with folks interested and hanging out with the coolest, brightest teammates I could ask for. So proud and happy to work with them, and definitely a highlight of my PhD.

Come check out Laughing Hyena today! Poster #441 at 10:45 arxiv: arxiv.org/abs/2310.18780 Michael Poli Dan Fu

Today we announce our collaboration with Capgemini to build next-generation AI solutions for enterprises. For the last months, we've been working on this together and now following Capgemini's participation in Liquid AI's successful $37.6m seed round, we are committed to

📢New research on mechanistic architecture design and scaling laws. - We perform the largest scaling laws analysis (500+ models, up to 7B) of beyond Transformer architectures to date - For the first time, we show that architecture performance on a set of isolated token

today, I want to share the core values that shape our culture at Liquid. here we go: no-bullshit meritocracy, burn the playbook, proactive execution and purposeful ownership, be white-box explainable and let's grow together. Allow me to elaborate: ------------- A CULTURE

scaling hypothesis holds strong! scaling laws have three components that work simultaneously with each other, and not alone: computation + algorithms + data. “hitting the wall” arguments around any of these components in isolation are bs.

.Liquid AI's new models are taking off 🛫 Great at English, Arabic, and Japanese

It has been said that AI is the new oil, the new electricity, and the new internet. And the once nimble and highly profitable software companies (MSFT, GOOG, ...) became like utilities, investing in nuclear energy, among other things, to run AI data centres. Open Source and the

In the shadow of DeepSeek R1, our LFM-7B is performing really well on OpenRouter. 🥳 Best-in-class 7B model at $0.01/M tokens (the same price as Llama 3.2 1B), low latency, and high throughput.

LFM2 is live. We keep state coefficients time-invariant because Hyena gated short convs already supply the adaptive dynamics—and the data agree. Same accuracy, leaner compute budget. Efficiency in practice, not on paper. #LiquidAI