Tessera
Folgen

Saul Kato

@focolab.bsky.social

I study things that think. Neuroscience professor at UCSF. Foundations of Cognition Lab. Reformed tech founder. www.focolab.org. https://substack.com/@saulkato

25 Posts150 Follower158 Folgt
Saul Kato@focolab.bsky.social · 5mo ago

HUGE congrats to my old Herophilus.com co-founder @seanescola.bsky.social on two fronts. The robotics company he co-founded was acquired by Amazon. apnews.com/article/amaz... He put out a paper proposing how to train large AI models so they acquire more cognitive functions arxiv.org/abs/2603.03414

Amazon buys Fauna Robotics, maker of the Sprout humanoid robot

Amazon has acquired Fauna Robotics, just under two months after the startup introduced a humanoid robot called Sprout designed for social spaces like homes and schools.

apnews.com

Saul Kato@focolab.bsky.social · 8mo ago

We noticed that a plate of C. elegans worms, under starvation conditions, forms stable living spatial structures, which we dub "worm webs". These webs seem to be scaffolded by an excreted substance from living or dead worms, which we dub "wormstuff". www.focolab.org/post/worm-we...

Worm Webs and Wormstuff

Sometimes a fresh pair of eyes in a lab notices a phenomenon that has been under the noses of researchers for a long time but ignored as a distraction. After more careful observations, replications, a...

focolab.org

Saul Kato@focolab.bsky.social · last yr.

Preprint posted for Kevin Zhou's awesome development in the Waller Lab of a gigapixel-scale microscope using a low-cost sensor array, along with Muneki Ikeda. www.focolab.org/post/large-s...

"Large-scale compressive microscopy via diffractive multiplexing across a sensor array" preprint released

Kevin Zhou, former postdoc in the collaborating lab of Laura Waller at Berkeley, and who recently started his own lab at University of Michigan, developed this way to achieve gigapixel-resolution high...

focolab.org

Saul Kato@focolab.bsky.social · last yr.

We built StarDist to detect cells in volumetric microscopy and jointly infer closed curved parametric surfaces for each cell. It's an example of an interpretable machine learning approach. More detail in the blog post and preprint. www.focolab.org/post/surfdis...

"SurfDist" pre-print posted

We developed an "interpretable machine learning" system to find and characterize blobs in volumetric images, such as microscopy images of cells, something we need to do all the time in our lab. The no...

focolab.org

Saul Kato@focolab.bsky.social · last yr.

An instant classic by Zheng and Meister ‪(@mameister4.bsky.social‬): www.cell.com/neuron/fullt... "...1⁢0^8... the ratio between peripheral information processing and the overall throughput of the brain. Numerically speaking, this may be the largest unexplained number in brain science. "

The unbearable slowness of being: Why do we live at 10 bits/s?

Zheng and Meister write about the paradoxical slowness of human behavior. Although our senses gather data at 109 bits/s, our overall information throughput is only 10 bits/s. This stark contrast touch...

cell.com

Saul Kato@focolab.bsky.social · last yr.

Data contamination (validation data leaking into training data) is a critical problem for the whole approach of (static) performance benchmarks to test large AI models. This issue needs to be surfaced more. This article for example, completely neglects it. www.nytimes.com/2025/01/23/t...

When A.I. Passes This Test, Look Out

The creators of a new test called “Humanity’s Last Exam” argue we may soon lose the ability to create tests hard enough for A.I. models.

nytimes.com

Saul Kato@focolab.bsky.social · last yr.

Prediction: AI will ultimately drive the renaissance of the fourth estate, the press, as we seek authoritative sources of information and trusted interpretation.

Saul Kato@focolab.bsky.social · 2y ago

Our paper on WormID is out. Moral: building a multi-lab data corpus boosts generalization and performance of ML labeling algorithms. We extended the Neurodata Without Borders format to support this. Resource at wormid.org. www.focolab.org/post/our-pap...

Our paper "Unifying community whole-brain imaging datasets enables robust neuron identification and reveals determinants of neuron position in C. elegans" is out.

https://www.cell.com/cell-reports-methods/fulltext/S2667-2375(24)00354-0Congrats Yutaka and the rest of the team! It has been a fantastic collaborative effort and we hope the community continues to bu...

focolab.org

Saul Kato@focolab.bsky.social · 2y ago

For our inaugural post on bsky, we're happy to share our preprint on tardigrades as a new model organism for neuroscience. Exciting times for these little critters! www.focolab.org/post/our-pre...

Our preprint "The tardigrade as an emerging model organism for systems neuroscience" is up.

We have been making the case for some time that tardigrades, aka water bears, would be an amazing new model organism for certain big systems neuroscience questions. We put this into a detailed paper o...

focolab.org