We removed 93.25% of the connections in our Un-0 image model, fully expecting to pay for it in quality. But it got better. FID 7.15 on ImageNet 64x64, roughly 1.9 ahead of the dense baseline at matched size. Same family of model, a fraction of the couplings, a better score. Here is why that is not as strange as it sounds. Un-0 is a coupled oscillator model, and in the dense version every oscillator talks to every other one. That sounds like a strength, but it means the whole system can fall into catastrophic synchronization: everything locks into step, gradients go flat, and learning stalls. Sparser connectivity leaves room for coherent and incoherent activity to coexist. The dynamics stay alive, and the model keeps learning. Connectivity turns out to be a control knob, not a dial you turn up until it stops. Learn more here:

Building a new memory means a twenty-year grind to drive the error rate toward zero. AI just made that grind optional. Because neural networks are remarkably tolerant of noise, we no longer have to chase punishing reliability targets. Training can settle into a good solution even when the hardware underneath is imperfect (which is exactly what the loss landscape below is showing). That single shift lets us stop optimizing for zero errors and start optimizing for what matters now: more bits per cell, higher density, lower energy. The universal memory never showed up, but AI may be the workload that tells each technology exactly which job it was born to do. Read the full breakdown by Giacomo Pedretti (MTS, AI Hardware Architectures) and Srenik Mehta (VP of Engineering):

What if your AI chip didn't calculate the answer, but physically settled into it? That's the premise behind dynamical system hardware, and there's never been a standard way to program it. In an upcoming International Symposium on Computer Architecture (@ISCAConfOrg) 2026 paper, Unconventional AI MTS Prof. Ang Li and co-authors introduce DS-ISA: a minimalist 9-instruction architecture that bridges digital processors and continuous-time analog physics. It's a proof-of-concept, but it's the abstraction layer the field needs to build real compilers and software stacks for dynamics-based AI hardware. Full deep dive:

Most companies use AI tools, but we are not a 'conventional company.' We're building Unconventional AI for this new age as an AI-native organization. In this latest blog, our [un]CFO, Ali Esfahani, shares how he has used AI in his first five months across finance, research, recruiting, legal, and other functions to help build a new kind of company. The result isn't just greater efficiency. It's a fundamentally different way of operating. Read more:

Lewis Hamilton recently pointed to a mismatch between simulation and reality as a major factor behind a challenging race weekend. It's a problem every systems engineer recognizes: how do you simulate a system operating at the edge of chaos? In our latest blog, we explore how Unconventional AI builds digital twins that faithfully model hardware operating in these highly sensitive regimes, capturing the physical and numerical effects that can significantly influence real-world performance. Read the full post:

Most real-world systems are dynamic. So why do we still treat computation as static? Our latest blog explores computation through motion using gyroscopes, rods, springs, and ordinary differential equations to perform handwritten digit classification. A deep dive into: • dynamical systems as compute • differentiable ODE solvers • physics-inspired machine learning • emergent computation through interaction Read here:

Tomorrow, May 15, is the final day to submit pre-proposals for the Unconventional Grant. Over the past several weeks, we’ve seen proposals spanning: • computation as dynamics • in-memory and in-physics compute • architectures that minimize data movement • new abstractions beyond linear algebra Many converge on the same intuition: meaningful efficiency gains in AI will not come from scaling existing approaches alone, but from fundamentally different ways of representing and computing. We are looking for technically grounded ideas that challenge assumptions across hardware, systems, and learning. We’re not looking for taller ladders to the moon. We’re looking for rockets.

Getting to 1000x energy efficiency in AI isn’t about one breakthrough. It’s about solving two hard constraints: 1. Data movement dominates energy 2. Amdahl’s Law caps system-level gains Which means you have to rethink everything: models, hardware, and how they’re designed together. If this kind of problem excites you, you’ll enjoy our latest blog:

Attending #ODSCAIEast# today? Join our cofounder @mcarbin at 2:35 PM ET in the Keynote Room. “Nonlinear Dynamics as the Next Substrate for Intelligence” explores a fundamental idea: we may be building AI on the wrong substrate. The human brain runs on ~20 watts. Today’s AI systems require orders of magnitude more. That gap isn’t just engineering, it’s architectural. Michael will share why moving beyond digital computation toward nonlinear physical dynamics could unlock a radically more efficient path to intelligence.

Analog vs. digital. Which is actually better? As AI systems hit efficiency limits, it might be time to rethink the abstractions we’ve taken for granted and focus on using the right tool for the job. Explore the case for a mixed-signal future in our latest blog post “Analog is dead, long live analog.”

We’re introducing the Unconventional Grant. A new research grant program supporting bold, unconventional ideas in AI. We’re allocating $500,000 in total funding, awarding up to five $100,000 grants to researchers exploring new paradigms in efficient, scalable, and biologically inspired AI systems. We’re especially interested in ideas that challenge how AI systems are built today, from unconventional circuits and architectures to new approaches in neural networks and theory. Not incremental work. Not safe bets. Ideas that push the field forward.

Stop copying the past. Is modern AI just a "cargo cult" worshipping the GPU? For 60 years, hardware and software have lived in completely separate worlds. But AI is forcing us to tear down those walls. The industry is stuck optimizing old abstractions and linear algebra not because it’s the best way to build intelligence, but because it’s what we inherited. We’re flipping the script at Unconventional AI and introducing neural co-evolution. We are building hardware and neural networks together from day zero to bypass the limits of traditional computing and unlock mind-blowing 1000x efficiency gains. The era of siloed design is over. Ready to see the future of compute? Read the full breakdown in our latest blog post.

In a recent episode of the @ThisWeeknAI, our CEO @NaveenGRao joined @jason for a roundtable discussion with @chaselochmiller (@CrusoeAI) and @ml_angelopoulos (@arena). Naveen discusses the long-term limits of scaling AI on today’s computing architectures and why new approaches to hardware and computation will be needed as the field continues to grow. The discussion also touches on infrastructure constraints, evaluation challenges, and how engineers can stand out as the industry matures. These are the kinds of questions we are exploring every day at Unconventional AI. Watch the full conversation below.

1/10 Reimaging computing using dynamical systems raises a host of fundamental questions, among them: How programmable/steerable is a candidate system? This week, we ran an experiment to test the limits of programmability by asking: is a toy 4-oscillator system expressive enough to dynamically sweep out any arbitrary pattern in phase-difference space? After testing our “[un]” logo, we concluded these systems are highly steerable. Here’s how we did it.