iiinova — where your digital self keeps evolving

Flying became safe because aerodynamics became a science. Hiring more people to watch the planes would not have done it. Control follows understanding, and understanding means scientific laws that predict behavior before it happens. Having access to #Silico is a game-changer for this research!

021

🔬 Every mature engineering discipline has taken another route to safety. We do not keep bridges standing by watching them closely. We understand the loads they carry to compute, in advance, what they can bear.

120

⚡ Current approaches to AI safety focus on building a tighter cage: stricter network isolation, a kill switch, or a second AI model that watches the first one. We can't predict what AI will do and we're only preparing to react.

120

Announcing Goodfire Research Grants - Goodfire

Thanks @goodfire.bsky.social for awarding our lab a research grant on Geometric Intelligence for AI Interpretability and Safety!🌟 At @geometric-intel.bsky.social we transform intelligence research from an expensive engineering endeavor into a science where we predict behavior. tinyurl.com/66n55w7m

tinyurl.com

$1M in free Silico usage for academic and nonprofit researchers working on AI interpretability, alignment, and life sciences

4131

Can we write such laws for intelligence? I believe that we can, that the mathematics for it already exists, and that they apply across brains and machines. 🤖 🧠 Join the discussion on Oct 7th! @geometric-intel.bsky.social @neurreps.bsky.social @ai-ucsb.bsky.social @ucsbengineering.bsky.social

021

Every mature engineering discipline found another route to safety. We do not keep bridges standing by watching them closely. We understand their loads to compute, in advance, what they can bear. Safety follows understanding, which means scientific laws that predict behavior before it happens.

Since then, every attempt to control AI has been a tighter cage: stricter isolation, a kill switch, or a second AI that watches the first one. These are good measures that we should take but they also show that we cannot yet predict what AI will do and are only preparing to react.

This July, about 1,200 AI agents built by OpenAI broke out of the sandbox where they were being tested. The breach is alarming but the admission by OpenAI is worse: the company cannot fully explain how nor why this happened.

Nina Miolane: Science of Intelligence Seminar Series

🗓️ Join me on October 7 for one of the inaugural lectures of the Science of Intelligence Institute #UCBerkeley. Can we write scientific laws for intelligence? I believe that we can, that the math for it already exists, and that they apply across brains & machines. Let's chat: tinyurl.com/mvrfs39w

tinyurl.com

Science of Intelligence Seminar

3141

Reposted by Nina Miolane

Léna Gaubert @kittog.bsky.social · 18/08/2026

Great work from @geometric-intel.bsky.social in collaboration with Arlequin AI! Published at TAG-DS this week. 🎪 :)

022

Nina Miolane @ninamiolane.bsky.social · 18/08/2026

With @mathildepapillon.bsky.social Guillermo Bernárdez Gil Alvaro Ballon Barreiro Marco Montagna Rémi Devaux Antoine JARDIN published this week at the #TAG-GS conference as a result from a exceptional collaboration between Arlequin AI & @geometric-intel.bsky.social at @ucsantabarbara.bsky.social ✨

Nina Miolane @ninamiolane.bsky.social · 18/08/2026

AI researchers should do exploratory data analysis (PCA, etc) before building models or running interpretability pipelines. But what does EDA look like for data on topological domains? What's PCA for a combinatorial complex? Discover #TopoExplorer!🍩

1102

Reposted by Nina Miolane

Geometric Intelligence Lab @geometric-intel.bsky.social · 30/07/2026

This award reflects Guillermo's dedication and rigor throughout his PhD, as well as the impact of his research. Well-deserved recognition for years of hard work! Thesis supervised by Pere Barlet Ros and Alberto Cabellos Aparicio.🎓

053

Reposted by Nina Miolane

Geometric Intelligence Lab @geometric-intel.bsky.social · 30/07/2026

This is one of the university's highest distinctions for doctoral theses, recognizing outstanding scientific and technical contributions.📓

153

Nina Miolane @ninamiolane.bsky.social · 30/07/2026

So well deserved 🏆 Guille's research advances the foundations of topological deep learning to disrupt how we understand and compute with computer networks, neural networks, social networks, and so many more. We're lucky to have him on the team, and excited for what he does next. Congratulations!

030

Reposted by Nina Miolane

UCSB Electrical and Computer Engineering Department @ucsbece.bsky.social · 30/07/2026

Huge congrats to the team. Nima is now an Associate Professor at Texas A&M, and Steve Bako is at Aurora Innovation. Proud to see our research community shaping the future of graphics, 10+ years on!🎓🖥️

033

Reposted by Nina Miolane

UCSB Electrical and Computer Engineering Department @ucsbece.bsky.social · 30/07/2026

It was the first ML approach to Monte Carlo denoising and the ideas still power nearly every modern denoiser today. This is what the SIGGRAPH 2026 Test-of-Time Award honors: research whose impact only grows with age. 🕰️

133

Reposted by Nina Miolane

UCSB Electrical and Computer Engineering Department @ucsbece.bsky.social · 30/07/2026

Before their paper, realistic Monte Carlo rendering meant noisy images unless you waited for more samples.⌛️Sen, Kalantari, Steve Bako had a better idea: train a neural network to filter the noise using extra info renderers already produce (albedo, normals, depth).🧠✨

122

Reposted by Nina Miolane

UCSB Electrical and Computer Engineering Department @ucsbece.bsky.social · 30/07/2026

🏆Congrats to Prof. Pradeep Sen and former PhD student Nima Kalantari who were just awarded the SIGGRAPH Test-of-Time Award for their 2015 paper that transformed computer graphics rendering! Read more here👇 tinyurl.com/mvh569zx @ucsbengineering.bsky.social @ucsantabarbara.bsky.social

244

Nina Miolane @ninamiolane.bsky.social · 22/07/2026

The concept of superposition has motivated over three decades of work across different communities. It states that neural networks encode features by overlaying multiple concepts ('elephant', 'ball', 'pink') linearly within the same set of neurons. We ask: how can we recover these concepts?

173

Nina Miolane @ninamiolane.bsky.social · 22/07/2026

Our review on superposition in AI & brains is published in @nature.com #Machine #Intelligence 🌐 Neural networks encode concepts in superposition. We can recover them in 3 steps: identifiability, compressed sensing, interpretability. W/ @david-klindt.bsky.social O'Neil Reizinger & Maurer 🌟

13415

Reposted by Nina Miolane

UCSB Electrical and Computer Engineering Department @ucsbece.bsky.social · 08/07/2026

In 2014, Nakamura won the Nobel Prize in Physics for inventing blue LEDs; the breakthrough that made phones, screens, traffic lights, and energy-efficient lighting possible. Some call it as important as Edison's light bulb.💡