Moving to another social media platform? I'm on the the mammoth one, the blue one, and the corporate one, if you want to follow me there (in order of preference). Links in next post.

Last chance to get your SNUFA spiking neural network abstracts in, deadline tomorrow. 300 words and a chance to give a talk in front of hundreds (plus YouTube upload gets hundreds more). snufa.net/2026/ Don't forget to register for the conf & tell your friends!

Very happy to have a role consulting with @CallosumAI. AI energy needs are growing out of control and access is controlled by too few US-based companies. Cheaper and more energy efficient AI can address these issues and open more responsible AI use more widely.

Today we're launching @CallosumAI out of stealth. I’m incredibly excited. We think AI & compute is about to change fundamentally. The current paradigm scaled up on an assumption that single models on identical chips would be enough. That shaped everything - models, data centres, the entire compute stack. It's taken us remarkably far. But what started as a practical bet has hardened into constraints now holding us back. The problems we need AI to solve - in science, industry, security - won't be solved by a single form of intelligence. They'll be solved by many. Intelligence is heterogeneous. Biology figured this out long ago. AI infrastructure hasn't caught up. At the same time, amazing new chips are arriving from every direction, each with radically different strengths. But every layer of the stack above them was built for homogeneity. But there is no infrastructure to compose diverse hardware and AI into Intelligent Systems that are greater than the sum of their parts. That's why Jascha Achterberg and I founded Callosum. Callosum is the Intelligent Systems Company. We are building the infrastructure where heterogeneous chips & intelligence co-evolve to solve the world's hardest problems. Our ambition is to open an entirely new paradigm, where diverse chips create algorithmic possibilities that didn't exist before, and new architectures reveal new ways to exploit the hardware. We've raised $10.25 million. Pre-seed led by @pluralplatform and @soundboy , with 22 world-class investors including @joinef, Charlie Songhurst, @nickmckeown1, @stanboland, @LawrenceLundy & supported by @ARIA_research. We've assembled an extraordinary team. I'm in awe of the people around me every day. Read about us in @FortuneMagazine: fortune.com/2026/02/26/start…

Two papers out, on a new paradigm of temporal computation! Our first work funded by @ARIA_research. We're super proud of this, and there's much more coming: Neural networks, specifically their weights, have become the most useful functional abstraction from the brain. As powerful function approximators they seeded the current era of AI. But there are many more useful abstractions. The brain does so much more than learn weights. At its core, the brain exploits the structure of the physical world to perform computation. It aligns itself to reality - to time and space. @achterbrain and I have long been working on how to embed neural networks in space and link this to hardware. But there's another dimension: time itself. We think leveraging both is a big part of the puzzle as to why human learning is so efficient. If this is true, how can we use space and time in neural networks? What would this even mean? We pitched to @BramhavarSuraj at @ARIA_research one possible way. Several years ago, I stumbled upon theoretical work that made it concrete. Time delays - the physical fact that signals take time to travel - can store memory (and other things, including increasing the number of computable functions). Even in feedforward networks. The delay here isn't overhead as would be traditionally thought but a feature that can be exploited. TL/DR - our main findings show that you can do computation in neural networks with time, without (much) need for weights. And it's remarkably efficient. We also show it's possible to co-design hardware (we open-source a chip design) with novel architectures that exploit time to maintain long contexts. In our first paper, we train neural networks to learn delays and weights. The result: state-of-the-art performance on all the temporally complex neuromorphic benchmarks we tested. Crucially, once you're encoding time directly, weights become almost irrelevant. We compress them to 1.58 bits, just positive, negative, or absent (ternary) weights. That's it. Model sizes drop to double-digit kilobytes! This works because we're finally encoding information the way the task needs it. Time and space. Which just so happens to be... everything in the physical world. Robotics. Embodied systems. Physical intelligence. In the second paper, we turn to memory. Intelligent systems don't just compute - they hold onto information over long context windows. And doing this efficiently in hardware underlying computation is a hard, unsolved problem. To solve this, we built a dual memory pathway architecture: fast spiking dynamics plus a compact state-space memory module that evolves much more slowly. Inspired by how the mammalian cortex separates fast somatic spiking from slower dendritic integration. Each layer maintains a tiny amount of working memory - just ~5% of hidden width - that summarises recent activity and feeds back into the network. There are many directions to scale this further. We co-designed the algorithms and hardware together from the ground up. The result: >4× throughput and >5× energy efficiency, beating Intel's Loihi2 and other leading neuromorphic platforms like DenRAM and ReckOn. We're open-sourcing the chip design so people can build on this. Incredibly proud of the team: @pengfeisun17, @achterbrain, @neuralreckoning, Zhe Su and @giacomoi One of our key takeaways: the current AI paradigm is a narrow slice of what's possible. Scaling homogeneous systems only gets you so far. Biological intelligence is deeply heterogeneous: different timescales, different substrates, different specialisations, all co-evolved together. We think the next frontier of scaling means embracing that heterogeneity. Algorithms and hardware aren't separate problems. They need to co-evolve together. Can't wait to share what we're cooking next.

Reminder if you missed #SNUFA spiking neural network and neuromorphic workshop earlier this month, all our talks were recorded and are now available to watch. piped.video/playlist?list=PL… #Neuroscience #ComputationalNeuroscience #SpikingNeuralNetworks

🚨 Fresh and alarming development in the Imperial pay dispute. The Joint Trade Unions have uncovered that key 2018 pay benchmarking recommendations—based on extensive staff consultation—were quietly set aside by College management. See here: ucu.imperial.ac.uk/archives/…

Talks from #SNUFA 2025 are now available on YouTube: piped.video/playlist?list=PL… #neuroscience #ComputationalNeuroscience #SpikingNeuralNetworks

#SNUFA day 2 starting in a few minutes. Lots of neuromorphic computing and SNNs. Check out snufa.net/2025 for agenda and links to join.

Day 2 of #SNUFA is starting in a couple of hours (1pm GMT). While yesterday was more classical comp neuro, today has a bit more of an engineering and neuromorphic focus. Hope to see you there!

We are live! Click agenda link below to join.

It's #SNUFA day! Starting at 1pm GMT (in 3h at the time of writing this). You can still register throughout the workshop, and click on the event page to get the zoom link to join.

Reminder this is happening this Wed/Thu. Free spiking neural network conference - registration required (see below).

Reminder everyone that #SNUFA is next week! Make sure to register in advance to get emailed the link to join.

Spiking NN fans - the #SNUFA workshop (Nov 5-6) agenda is finalised and online now. Make sure to register (free) soon. (Note you can register for either day and come to both.) Agenda: snufa.net/2025/ Registration: eventbrite.co.uk/e/snufa-202… Thanks to all who voted!

Dear @ElsevierConnect, having a default opt-in to marketing emails is in breach of the GDPR. Please correct this.

Message for participants of the #SNUFA spiking neural network workshop. We got ~60 awesome abstract submissions, and we'd now like your help to select which ones should be offered talks. To take part, follow the "abstract voting" link at: snufa.net/2025/ Thanks! ❤️

Today is the last day for submissions to the free online SNUFA spiking neural network workshop! Great opportunity to present work (100s of participants and sometimes 1000s of YouTube views per talk). Abstract submission is easy (300 words). Do it now! snufa.net/2025/

In our latest “This paper changed my life,” @neuralreckoning explains how Friedemann Zenke’s @hisspikeness 2019 paper, and its related coding tutorial SpyTorch, made it possible to apply modern machine learning to spiking neural networks. thetransmitter.org/this-pape…