iiinova — where your digital self keeps evolving

Reposted by Gabriel Béna 🌻

Dan Goodman @neural-reckoning.org · 01/09/2026

If you're in or near London do think about coming along to this. Not just a talk about a really fun topic (neural heterogeneity) but a chance to grill the author (me) on whether we learn anything with the theoretical methods he's used, and indeed whether the questions he poses are even meaningful.

1154

Gabriel Béna 🌻 @solarpunkgabs.bsky.social · 19/08/2026

Huge ❤️ to my co-author Marcello Barylli, and to Alexander Mordvintsev (who dreamt up the original differentiable circuits), Eleni Nisioti, and @risi.bsky.social. 📄 Self-Organising Digital Circuits, #ALIFE2026 (in press today) 🔗 self-organising-circuits.github.io/ #NCA #ALife #SelfOrg

self-organising-circuits.github.io

Self-Organising Digital Circuits

Extending the Neural Cellular Automata paradigm from pattern formation on grids to functional logic generation and self-repair on arbitrary graphs.

000

Gabriel Béna 🌻 @solarpunkgabs.bsky.social · 19/08/2026

I'm presenting all of this TOMORROW at #ALIFE2026 — if you're here, come heckle 🎤 And the long game? Self-healing hardware: FPGAs, neuromorphic chips… maybe something that keeps computing on a deep-space probe no one can reboot. 🛰️

100

Gabriel Béna 🌻 @solarpunkgabs.bsky.social · 19/08/2026

Enough talk — go break it yourself 🎮 The demo runs the real trained model in your browser (no server, no cap). Click a gate to kill it. Hit Shotgun/Shuffle for chaos. Watch the truth tables rewire live and the accuracy (hopefully) crawl back.

100

Gabriel Béna 🌻 @solarpunkgabs.bsky.social · 19/08/2026

Why care? Because damage, device drift, and a brand-new task are secretly the same problem: "the current config stopped working — find another that does." One local rule that hops between working configs = a compiler + a repair crew + a lifelong learner, all in one. 🔧🩹📚

110

Gabriel Béna 🌻 @solarpunkgabs.bsky.social · 19/08/2026

We mapped 50,000 healed circuits — all computing the same function, all perfect. SODC (left): a galaxy of distinct solution "archetypes”, a whole neutral landscape. Backprop (right): one lonely blob. Degeneracy, visualized.

Two UMAP scatter plots embedding 50,000 circuit configurations that all compute the same function perfectly (UMAP dimension 1 on x, dimension 2 on y). Points are coloured by how many perturbation–recovery cycles they are from the root circuit, which is marked with a red star; thin grey lines link successive states. Left ("TMT", the local policy): the points organise into many separate clusters spread across the whole space — a landscape of distinct structural "archetypes". Right ("BP", backpropagation): a single dense blob around the root. Same task, same accuracy — but the local policy explores a far more degenerate solution space.

110

Gabriel Béna 🌻 @solarpunkgabs.bsky.social · 19/08/2026

And also, it's scale-free in the best way: train on tiny 264-gate circuits, deploy on 450+ gates it's never seen… and accuracy goes UP 📈 Turns out "learn a local rule" generalizes way better than "memorize a big circuit." Who knew. (biology knew.)

Two line charts of hard accuracy (y-axis, 0.5–1.0) versus circuit width in number of nodes (x-axis, ~170–450). A dashed vertical line marks the training size of 264 nodes; green is evaluated without damage, blue with damage. Left ("fixed wiring"): accuracy spikes to ~1.0 exactly at the 264-node training size and collapses to about 0.5 on either side — the policy overfits the scale it trained on. Right ("random wiring"): accuracy stays high (~0.8–0.9) across all widths and even rises as the circuit grows past the training size, with shaded confidence bands — scale-freedom is a learned capability.

110

Gabriel Béna 🌻 @solarpunkgabs.bsky.social · 19/08/2026

So naturally, we shot the circuit. Repeatedly. 🔫 The wild part: when a gate dies, SODC does not restore the original. It hops to a different configuration that computes the same function. Repair ≠ rewind. Repair = find another way.

A grid of PCA trajectory plots — principal component 1 on the x-axis, component 2 on the y-axis — tracing how a circuit's configuration moves while the local policy repairs it. Line colour runs from dark blue at step 0 to yellow at step ~250. Columns: no damage, stochastic failures, shotgun failures; rows: recoverable vs permanent damage. With no damage the trajectory is a single clean arc. Under damage the paths fan out and settle in different places — the policy re-routes to new, functionally-equivalent configurations instead of rewinding to the original.

110

Gabriel Béna 🌻 @solarpunkgabs.bsky.social · 19/08/2026

This change flips everything: A target pattern admits ONE answer. A target function admits MANY. Every wiring that computes f is equally "correct" → the solution space is drenched in degeneracy. The exact thing biology leans on. We didn't add it. We just stopped forbidding it.

110

Gabriel Béna 🌻 @solarpunkgabs.bsky.social · 19/08/2026

The move: make every cell a logic gate: its state IS the truth table (a small look-up table). The NCA is no longer the computer. It's the tiny, decentralized optimizer that assembles + repairs the circuit, one local message at a time. No global backprop at deploy — just forward passes. ⚡

110

Gabriel Béna 🌻 @solarpunkgabs.bsky.social · 19/08/2026

Here's my whole NCA arc in one breath: 🖼️ NCA: grow a pattern (answer lives in the pixels) 🧮 UNCA: compute many functions (but the answer still lives in the cells) 🔌 SODC: the cells stop holding the answer and start emulating the computation! gabrielbena.github.io/blog/2025/b...

gabrielbena.github.io

A Path to Universal Neural Cellular Automata

Exploring how neural cellular automata can develop continuous universal computation through training by gradient descent

110

Gabriel Béna 🌻 @solarpunkgabs.bsky.social · 19/08/2026

Quick recall: an NCA = a grid of cells, each seeing only its neighbours, all running the SAME learned rule. None told its job, yet they specialise: an eye here, a tail there. A lizard 🦎 grows from one pixel; halve it, it regrows. grow → persist → repair distill.pub/2020/growin...

distill.pub

Growing Neural Cellular Automata

Training an end-to-end differentiable, self-organising cellular automata model of morphogenesis, able to both grow and regenerate specific patterns.

121

Gabriel Béna 🌻 @solarpunkgabs.bsky.social · 19/08/2026

Biology refuses to play that game. Lose a chunk of tissue and the brain doesn't reach for a backup: it reroutes, recovering the same function through different parts. The trick has a name: degeneracy: many different structures, one same function. 🧬 (hold that thought)

110

Gabriel Béna 🌻 @solarpunkgabs.bsky.social · 19/08/2026

How does today's hardware survive faults? Brilliantly, but by anticipating them: redundant cores, spare memory rows, error-correcting codes. Decades of impressive engineering. The catch: it only covers failures you modelled in advance. Go past the budget and it breaks — hard. 💥

110

Gabriel Béna 🌻 @solarpunkgabs.bsky.social · 19/08/2026

🤖 🧠 🧪 New paper! I'm presenting it TOMORROW at #ALIFE2026! What if a chip could heal itself the way a brain does: not with a spare part, but by rewiring around the wound? Meet Self-Organising Digital Circuits. 👉 self-organising-circuits.github.io/ (go murder some logic gates, I'll explain) 🧵⬇️

self-organising-circuits.github.io

Self-Organising Digital Circuits

Extending the Neural Cellular Automata paradigm from pattern formation on grids to functional logic generation and self-repair on arbitrary graphs.

3237

Reposted by Gabriel Béna 🌻

Marcus Ghosh @marcusghosh.bsky.social · 14/08/2026

Now published @commsaicomp.nature.com with @neural-reckoning.org We: 🤖 Define a taxonomy of neural network architectures (here 128 unique structures) 🦾 Train > 25,000 models on different navigation tasks #RL 🧠 Link each architecture's behaviour to its memory dynamics doi.org/10.1038/s444...

A taxonomy of recurrence. We defined a family of neural networks for exploring questions in neuroscience and machine learning. Here each network is represented by three circles joined by different connections (short, coloured lines). Networks are joined by longer, pale grey lines if they differ by one connection. Moving from left to right takes us from the simplest to the most complex network.

13411

Reposted by Gabriel Béna 🌻

WeRateDogs @weratedogs.com · 09/07/2026

This is Oreo. He will eventually eat the blueberry. But first he must establish dominance over it. 12/10 (TT: oreominibernedoodle)

452142121760

Reposted by Gabriel Béna 🌻

Wcrt L @l-wcrdt.bsky.social · 08/05/2026

Le 8 mai est antifasciste. Le 8 mai, c'est la victoire contre l'extrême droite.

Réponse au tweet de Marine Le Pen célébrant le 8 mai 1945 :

"Bonjour, votre parti a été fondé par des SS.

La victoire du 8 mai 1945 n'est pas la vôtre, bien au contraire".

1319976

Gabriel Béna 🌻 @solarpunkgabs.bsky.social · 27/04/2026

Stop posting AI slop please you're not helping

000

Gabriel Béna 🌻 @solarpunkgabs.bsky.social · 26/04/2026

Wait I don't even understand their reality?? What's the argument and link between this pseudo gunman and the ballroom and how would one legitimize the other, am I daft or is maga fucked up sens of reality becoming too twisted to even fathom?

000

Gabriel Béna 🌻 @solarpunkgabs.bsky.social · 12/04/2026

Qobuz is great, while Deezer is actually not so much better than spot...

000

Gabriel Béna 🌻 @solarpunkgabs.bsky.social · 12/04/2026

You're directly responsible for the rise of the far right across Europe you neoliberal twat

010

Reposted by Gabriel Béna 🌻

Bon Pote @bonpote.com · 06/04/2026

La pétition contre la loi Yadan vient d'atteindre les 400 000 signatures ! à 500 000, l’Assemblée nationale pourra décider de débattre de la pétition en Commission... ET dans l’hémicycle ! A partager un maximum : petitions.assemblee-nationale.fr/initiatives/...

22817630

Gabriel Béna 🌻 @solarpunkgabs.bsky.social · 24/03/2026

Bah si ils parlent du fait que pour la majorité de la "gauche" molle la lutte contre l'antisémitisme est devenu une arme politique plutôt qu'un vrai combat à mener, c'est plutôt vrai. Mais j'ai comme l'impression que ce n'est pas ce qu'ils racontent

160

Reposted by Gabriel Béna 🌻

Sébastien Fontenelle @vivelefeu.bsky.social · 17/03/2026

Communiqué de Sébastien Delogu: «Face à l'irresponsabilité d'un homme, nous serons responsables pour un million. Nous retirons notre liste afin de ne pas cautionner une stratégie qui fait courir un risque grave à notre ville.» Honte à toi, @partisocialiste.bsky.social.

7597187

Reposted by Gabriel Béna 🌻

Attac France @attac.org · 27/02/2026

Face à la montée des tensions lors des rassemblements pour le 8 mars, nous, syndicats et associations, demandons au ministre de l’intérieur Laurent Nuñez d’interdire la présence du collectif Némésis dans nos manifestations. france.attac.org/se-mobiliser...

france.attac.org

Lettre ouverte à Laurent Nuñez : Nous demandons l’interdiction de la présence de Némésis aux manifestations du 8 mars

Face à la montée des tensions lors des rassemblements pour la journée internationale des droits des femmes, nous, syndicats et associations, alertons sur les risques de débordements liés à la présence...

2197109

Reposted by Gabriel Béna 🌻

Kat Abughazaleh @katmabu.bsky.social · 26/02/2026

Tax billionaires out of existence.

546168293782

Reposted by Gabriel Béna 🌻

Thomas Lemahieu @thomaslemahieu.bsky.social · 22/02/2026

« On peut être deux, trois filles à tracter là où vous voulez les choper, un peu pour faire l’appât. » Révélations @humanite.fr sur les conversations secrètes, à Lyon, entre Némésis et les néofascistes pour tendre des pièges à la gauche et aux antifascistes. www.humanite.fr/politique/an...

humanite.fr

Dans les conversations secrètes entre Némésis et des néonazis pour agresser des militants de gauche à Lyon - L'Humanité

À Lyon, dans des échanges internes datant de l’automne dernier, que l’Humanité dévoile aujourd’hui, les fémonationalistes et d’autres groupuscules d’extrême droite planifiai...

3213001034

Reposted by Gabriel Béna 🌻

Sébastien Fontenelle @vivelefeu.bsky.social · 22/02/2026

Au lendemain d’une manifestation néonazie, Jérôme Guedj dénonce la «responsabilité morale» de La France insoumise et Yannick Jadot déclare que «Jean-Luc Mélenchon doit faire les clarifications nécessaires pour rester dans l’arc républicain». #OnVousVoit.

2733194

Reposted by Gabriel Béna 🌻

Ben Collins @bencollins.bsky.social · 21/02/2026

This is a really good debate format for like 50 different reasons, but especially at this moment. There's a lot of equivocating over stuff that is not that complicated.

803171454

Reposted by Gabriel Béna 🌻

Sébastien Fontenelle @vivelefeu.bsky.social · 20/02/2026

Pour un front unitaire antifasciste.

3263137

Reposted by Gabriel Béna 🌻

Attac France @attac.org · 12/02/2026

⚠️ Scandale Amélie de Montchalin a été nommée première présidente de la Cour des comptes par Emmanuel Macron. Signez la lettre ouverte au président de la République pour exiger le retrait de cette nomination qui piétine la démocratie et l'Etat de droit👇 france.attac.org/actus-et-med...

france.attac.org

[Je signe ✍️] Amélie de Montchalin, de Bercy à la Cour des comptes, c'est non !

Ce mercredi 11 février, Amélie de Montchalin a été nommée première présidente de la Cour des comptes par Emmanuel Macron. Signez notre lettre ouverte au président de la République pour exiger le retra...

25246187

Gabriel Béna 🌻 @solarpunkgabs.bsky.social · 29/01/2026

Ok je vois, dur à saisir quand on ne maîtrise pas le sujet. Merci pour votre travail 😊

000

Gabriel Béna 🌻 @solarpunkgabs.bsky.social · 29/01/2026

Super article merci. Comment est ce que ces conclusions affectent les assurance vie "vertes" type goodvest / greengot ??

100

Reposted by Gabriel Béna 🌻

Mark D Humphries @markdhumphries.bsky.social · 30/12/2025

Just published my review of neuroscience in 2025, on The Spike. The 10th of these, would you believe? This year we have foundation models, breakthroughs in using light to understand the brain, a gene therapy, and more Enjoy! medium.com/the-spike/20...

medium.com

2025: A Review of the Year in Neuroscience

Enlightening the brain

614366

Reposted by Gabriel Béna 🌻

CMC Unit @cmc-unit.bsky.social · 14/01/2026

For the first Learning Club @cmc-lab.bsky.social in 2026, @marcusghosh.bsky.social on Jan 15, Thu, 2pm CET, will tell us about his recent paper (👇) [joint work w/ @neural-reckoning.org]. Want to attend, send an empty email to virtual talk-link-request@cmclab.org to get the link!

0105

Reposted by Gabriel Béna 🌻

Marcus Ghosh @marcusghosh.bsky.social · 01/08/2025

How does the structure of a neural circuit shape its function? @neuralreckoning.bsky.social & I explore this in our new preprint: doi.org/10.1101/2025... 🤖🧠🧪 🧵1/9

A diagram showing 128 neural network architectures.

711340

Reposted by Gabriel Béna 🌻

Kat Abughazaleh @katmabu.bsky.social · 07/01/2026

Our first ad of 2026! Because politics stuck in 2006 won't meet the moment. It's time to stand up to ICE, white supremacists, and the billionaire oligarchy.

18668801694