iiinova — where your digital self keeps evolving

Reposted by Jan Lause

Scott Marek @smarek0502.bsky.social · 11/06/2026

What matters most for childhood brain organization? We analyzed 649 variables. The answer: Socioeconomics (SES); with brain patterns pointing at sleep & stress as drivers. Even brain-IQ associations were better explained by SES. In Science today: www.science.org/doi/10.1126/...

6319169

Jan Lause @janlause.bsky.social · 12/06/2026

thank you! ❤️ great to have you and your team with us!

010

Reposted by Jan Lause

Marla Feller @mfeller.bsky.social · 12/06/2026

what an incredible achievement and resource for retina researchers. We use it, we love it — it is a game changer for all of us who love retinal circuits. Kudos.

static.klipy.com

Journey from Eye to Galaxy for National Space Day

ALT: Journey from Eye to Galaxy for National Space Day

284

Reposted by Jan Lause

ML for Science @ml4science.bsky.social · 11/06/2026

Eyewire II is a great research project from @philipp.hertie.ai's lab and international collaborators! Read @janlause.bsky.social's thread to find out more👇

096

Reposted by Jan Lause

Michael Okun @michael-okun.bsky.social · 10/06/2026

I’ve officially resigned as Associate Editor for Frontiers in Systems Neuroscience. It used to be a reputable journal, but became a case study in how forced automation destroys academic integrity. 👇

5435861430

Reposted by Jan Lause

Jan Lause @janlause.bsky.social · 09/06/2026

Meet Eyewire II: a new connectomic resource for the mouse retina. ~1 mm² of retina at nanometer resolution, with synapses and circuits of ca. 100,000 neurons, plus visual responses from the ~400 neurons shown in the video! Preprint: doi.org/10.64898/202... Data: eyewire.ai 🧠📈 🧠💻🧪 #VisionScience

25221

Jan Lause @janlause.bsky.social · 09/06/2026

2/2 Re: Spacing & type connectivity: We have not yet analyzed ribbon spacing, but did some type connectivity in a small subvolume - attaching Figure 8 from the resource paper that contains the results, and pinging @simoneebert.bsky.social who did the analysis :)

Scientific figure with panels A-G, with some neuron morphologies on top, zoom in on a synapse (with electron microscopy), and some connectivity and ribbon statistics below.

(A-E) Synaptic connectivity analysis exemplified in the rod pathway. Cross section (A) showing a ribbon synapse (open arrow) in a rod bipolar cell (RBC) axon terminal (green), a conventional synapse (filled arrow) in an A17 amacrine cell (AC) process (light blue) providing reciprocal feedback to the RBC, and an A2 AC process (in dark blue). 3D render of same synapse (B) and of involved cells (C). (D) Outputs (synaptic ribbons) and inputs (conventional synapses) of RBCs sorted by AC type (data from n=9 RBCs, each dot represents one cell pair; MF, medium-field). (E) Example skeletons of main AC types that synapse with RBCs (E), with Off and On starburst amacrine cell (SAC) stratification indicated (cyan and apricot, respectively). (F,G) Quantification of automatically detected ribbons in different bipolar cell (BC) types, with ribbon count (F, left), mean ribbon size (F, right), and distribution across IPL depth in (G, cyan and apricot lines indicate On and Off SAC plexuses, respectively; same BC dataset as in Fig. 4).

010

Jan Lause @janlause.bsky.social · 09/06/2026

1/2 Synapse prediction is still ongoing. We have a detector for synaptic ribbons (i.e., the excitatory presynapse) and found 6.5 million in the whole IPL. Assuming it is complete in the ca. 1x1mm piece, one could derive ribbons-per-cubic-micron or so. For the rest we would need the postsyn. partners

120

Jan Lause @janlause.bsky.social · 09/06/2026

🥲 too bad - so where are you getting your science news these days?

110

Jan Lause @janlause.bsky.social · 09/06/2026

Eyewire II is based in @unituebingen.bsky.social @princetonneuro.bsky.social and @feinberg.northwestern.edu, and funded by @erc.europa.eu , Research to Prevent Blindness and the Hertie foundation.

Funders logos: Research to prevent blindness, European Research Council, Brain Connects, Brain Initiative

120

Jan Lause @janlause.bsky.social · 09/06/2026

This is just the beginning. We're excited to see the discoveries that emerge from Eyewire II and to continue building this platform together with the retina community. See you at the #FASEB retina meeting this week @faseborg.bsky.social, and stay tuned for our next Eyewire Town Hall in fall!

111

Jan Lause @janlause.bsky.social · 09/06/2026

Eyewire II continues the tradition of public participation in connectomics: Citizen scientists helped proofread and annotate neurons alongside researchers and specialists - shoutout to Nikitas Serafetinidis (named coauthor on the paper) and to all other volunteers contributing!

220

Jan Lause @janlause.bsky.social · 09/06/2026

New to Eyewire and want to get started? Check out our website for ⚒️ Tools - eyewire.ai/tools - from command line tools to interactive data explorers 🦠 Cells - eyewire.ai/cells - direct links to all cells identified so far 🧭 Guides - eyewire.ai/guides - tutorials for a smooth start

Screenshot of https://eyewire.ai/tools, listing tools for Eyewire II, like Flatone, Morphotyper, Skeliner, .... and more

Screenshot of the morphotyper interface, with the F mini On RGC type selected. Cell selection menu on the top left, type stratification overview on the top right, mosaic view of all cells in the bottom right, showing that the cells tile the area shown regularly.

242

Jan Lause @janlause.bsky.social · 09/06/2026

Eyewire II is more than a dataset, it’s a community effort! Already now, >80 researchers from >30 labs are part of the Eyewire II, contributing their expertise to the consortium work. Meet them here eyewire.ai/consortium Join us today! eyewire.ai/ew2_access

Screenshot of two zoom panels, with many portrait views of people on the call

130

Jan Lause @janlause.bsky.social · 09/06/2026

But how do these cell types connect? Synapse prediction is ongoing, but Eyewire II already includes automatic detection of synaptic ribbons - the hallmark of retinal excitatory synapses. We auto-detected ~6.5 million ribbons so far.

110

Jan Lause @janlause.bsky.social · 09/06/2026

We built a human-in-the-loop approach: the model proposes types, human experts refine and correct. As a proof of principle, we applied it to bipolar cells — and recovered all 15 known types. See those and more types in our gallery 👇 📸 eyewire.ai/gallery

15 images on a grid, each showing a single side views of different BC type examples. Cells sit between two nuclei layers (GCL and INL), and extend their axon terminals down. The differences between the types' stratification is clearly visible.

Similar to the image before, but this time for 12 RGC type examples. Some types are asymmetric, others stratify broadly, others densely in the IPL.

121

Jan Lause @janlause.bsky.social · 09/06/2026

First of all, we need to catalogue those cells into types - to get to a “parts list” of the circuits that they form. Often, we can tell the type of a cell by its shape - but with this many cells, sorting them by hand doesn't scale. So we type them automatically, using machine learning.

Scientific figure with three panels A/B/C. The center is a colorful t-SNE embedding of Eyewire II cell morphologies. Single islands are highlighted and example morphologies from those islands are shown around the t-SNE map.

Full caption:
(A) Illustration of the processing pipeline. From left to right: The EM stack was segmented using a previously developed algorithm. The automatically segmented cell meshes were proofread by professional proofreaders and researchers. We computed skeletons from the meshes using skeliner, which were then flattened using pywarper. From the flattened skeletons, we computed morphological features, such as density profiles. (B) Normalized densities along z-axis across the inner retina for all 59, 000 neurons split by cell class: retinal ganglion cells (RGCs), amacrine cells (ACs), and bipolar cells (BCs). Cyan and apricot-colored lines indicate stratification levels of Off and On starburst cells (SACs), respectively. (C) t-SNE computed from 43 normalized morphological features of all proofread RGCs (red), ACs (blue; except On SACs, apricot, and Off SACs, cyan), and a subset of 200 BCs per type (green), with representative example neurons in x-z and x-y view, as well as cell distribution (dashed outline, outline of IPL; manually annotated cells, black dots; additionally automatically classified cells, blue dots).

150

Jan Lause @janlause.bsky.social · 09/06/2026

After 1 year of proofreading, Eyewire II already includes: • >8,000 bipolar cells • >13,000 amacrine cells • >4,000 retinal ganglion cells …and that is only 25% of all expected neurons in the data!

130

Jan Lause @janlause.bsky.social · 09/06/2026

The retina is one of the best systems for understanding how neural circuits perform computations. But until now, retinal connectomics datasets have been too small to capture rare cell types and neurons with large dendritic arbors. Eyewire II changes that - its 10x larger than previous datasets!

Different EM dataset sizes shown as colored rectangles with cell skeletons inside.
Eyewire II (dark blue, containing a wide field amacrine cell) covers almost 1 mm2, ca. 10x larger area than the earlier Eyewire I dataset (cyan, contains a SAC or sONalpha RGC), and ca. 100x larger area than e2006 (gray, contains an XBC).

110

Jan Lause @janlause.bsky.social · 09/06/2026

25221

Jan Lause @janlause.bsky.social · 08/06/2026

1-2-1-2 *taps mic* Is this still working?!

310

Reposted by Jan Lause

Trish Greenhalgh @trishgreenhalgh.bsky.social · 21/01/2026

As the world hurtles towards geopolitical crisis, my marine biologist son sends photo of an underwater leaf slug from Fiji to wish us calm.

2045767

Jan Lause @janlause.bsky.social · 21/01/2026

Very nice tool to test this! For a few original texts (that just live on my private devices and has never been uploaded anywhere) I could not get the de-BLANK-ing to work at all though... Maybe my excerpts where just too short or special-domain...

000

Reposted by Jan Lause

David Picard @davidpicard.eurosky.social · 20/01/2026

Coming soon to all of Europe? I hope so. I've been complaining about our lack of autonomy in the domain for as long as I've been using Linux.

1232

Reposted by Jan Lause

Brian Tyler Cohen @briantylercohen.bsky.social · 17/01/2026

More than 200,000 Danish citizens have signed a petition to buy California as a response to Trump’s attempt to take Greenland. They say they will provide Californians with “rule of law, universal health care, fact-based politics, and a lifetime supply of Danish pastries.”

1244187104992

Jan Lause @janlause.bsky.social · 16/01/2026

Want a relaxing break? Scroll this thread instead of your timeline. 🐈 (For best results, close this app once you are done and check your own surroundings for cats you don’t own — maybe you get lucky!

020

Jan Lause @janlause.bsky.social · 16/01/2026

Comment from Germany (where all of this might be different) — but is there a chance that these differences arise from how close each generation lives to other relatives like grandparents/uncles&aunts/…? Maybe boomer care time is lower because grandparents where closer to help out more?

1120

Jan Lause @janlause.bsky.social · 14/01/2026

Terns are my favorite type of bird :) Thanks for posting this overview!

010

Reposted by Jan Lause

Andrew B. Watkins @windjunky.bsky.social · 14/01/2026

Globally, no single day in 2025 was cooler than its 1991-2020 average. climate.copernicus.eu/global-clima...

211008674

Jan Lause @janlause.bsky.social · 14/01/2026

so nice! What a pretty bird 🐦

010

Jan Lause @janlause.bsky.social · 13/01/2026

The numbers in this overview were pretty surprising to me - I wonder how they look for Germany... (were we also suffer from low economic mobility etc...)

000

Reposted by Jan Lause

Tomer Ullman @tomerullman.bsky.social · 09/01/2026

once again being driven insane by ML conference submissions

311118

Reposted by Jan Lause

Jonathan Birch @birchlse.bsky.social · 08/01/2026

Martin Peterson's creative response to being banned from teaching Plato (shared with his permission).

An email from Martin Peterson to university administrators.

6150991520

Jan Lause @janlause.bsky.social · 08/01/2026

Interesting work on LLM memorization: Researchers were able to extract long book passages verbatim from Claude, Gemini and Grok --- sometimes almost the whole book. Curious to see how this type of result influences Copyright cases... Also see their thread with details and limitations below ⬇️

040

Reposted by Jan Lause

Altmetric @altmetric.com · 08/01/2026

Since Jan 1 2025, which feels like four trillion years ago, research has been shared on here 5 million whole-ass times. Bluesky recently passed 2 billion posts IN TOTAL. So 0.25% of the entire site's traffic was citations to research. That is actually massively high. Is it? Yes. Here's why.

Altmetric Explorer screenshot: Showing 5,011,088 mentions (from 4,918,333 individual posts) of research outputs from the results of your search query.

9299111

Reposted by Jan Lause

Christian Wolf @chriswolfvision.bsky.social · 06/01/2026

I have been asked several times how we draw the 3D Figures for of our papers, so I wrote a blog post on it. This does not replace a Blender tutorial, just a couple specifics for scientific papers: - import images as planes - render edges - transparent BG chriswolfvision.medium.com/creating-3d-...

15716

Jan Lause @janlause.bsky.social · 06/01/2026

cool map! What is happening in the black areas in the south?

110

Reposted by Jan Lause

Ruben C. Arslan @ruben.the100.ci · 05/01/2026

lol. Before Stackoverflow I knew SelfHTML and R-Help. Stackoverflow et al. were so welcoming and well-designed by comparison (e.g., they would search existing posts based on the text you wrote). Meanwhile, all my current LLM prompts include stuff like "please please tell me when my premise is false"

1151

Reposted by Jan Lause

David Mimno @dmimno.bsky.social · 04/01/2026

From the 2022 drop era: “who needs wheat? Bread is so much tastier”

1223

Reposted by Jan Lause

John F Wu @jwuphysics.bsky.social · 04/01/2026

To include some numbers here, StackOverflow in Dec 2025 had 3862 new questions. The last time SO had this few questions was in *Aug 2008*, the first full month that SO was in operations! In its peak month (March 2014), SO had 207,493 new questions!

082

Reposted by Jan Lause

Annika Brockschmidt @ardenthistorian.bsky.social · 03/01/2026

Because international media largely seems to struggle to choose the correct wording to describe what’s happening: the US, under the self describes „peace president“, attacked another country and kidnapped its president. That is factual, no matter how bad or illegitimate Maduro‘s rule is (very bad).

422607790

Reposted by Jan Lause

Aaron W. Gordon @agordon.me · 02/01/2026

One thing I wrote about here is how technology has made buses *way* more convenient while techno-futurists were obsessed with flying taxis and self-driving cars.

162214559

Jan Lause @janlause.bsky.social · 19/12/2025

Warum brauchts Journalismus von Netzpolitik.org? Nur so erfährt die Öffentlichkeit von Datenleaks und problematischen Entwicklungen in Richtung Überwachungsstaat (Palantir, Chatkontrolle..). Digitale Zukunft geht nicht ohne unabhänge Medien, die sich auskennen! Also, wer kann: Jetzt spenden!

netzpolitik.org

netzpolitik.org - Online-Medium für digitale Freiheitsrechte

042

Jan Lause @janlause.bsky.social · 19/12/2025

not a single query estimate, but just saw this paper from 2 days ago in my feed. The attached thread also puts it in perspective a bit.. bsky.app/profile/glob...

000

Jan Lause @janlause.bsky.social · 19/12/2025

Thanks, good to know it comes from OpenAI directly. I am afraid they have an interest to downplay this number so I would take it with a grain of salt.. the article puts it in context nicely, saying that its unclear what an average query is and how to account for the huuuge initial training costs

010

Jan Lause @janlause.bsky.social · 18/12/2025

Where is the estimate from? I have been looking for a good source for this for a long long while....

110