Kyle
banner
darsnack.bsky.social
Kyle
@darsnack.bsky.social
610 followers 500 following 89 posts
NeuroAI Scholar @ CSHL https://www.darsnack.info 
 Previously maintaining FluxML to procrastinate

Previously EE PhD at UW-Madison, comp. eng. / math at Rose-Hulman
Posts Media Videos Starter Packs
Personal bias but I liked this transmitter article: www.thetransmitter.org/systems-neur...

“You won’t catch an engineer trying to formulate a grand unified theory of the brain”

“Engineers know how to make things work […] This is what makes engineering closer to biology than to other […] sciences”
Neuroscience needs engineers—for more reasons than you think
Adopting an engineering mindset will help the field focus its research priorities.
www.thetransmitter.org
Reposted by Kyle
After 13 years in the US, I’ve made the difficult decision to leave. Having packed up everything and rethought about priorities, rather painstakingly, while I’m sad to leave the life I’ve made here, I’m also relieved that I won’t have to plan my life around immigration policies anymore.
Reposted by Kyle
I wish the whole unicode set was available from every emoji picker.
Reposted by Kyle
We're hiring in Seattle! HMU if you're affected by Amazon layoffs and have #julialang experience
Reposted by Kyle
Is this what Cuo-mentum looks like? How close is this race? And how can we even predict an unprecedented competitive general election mayoral race when New York City hasn't had one since 2009?

Dear reader, we have the answers at this week's edition of MAYORAL SPEW.

hellgatenyc.com/oh-god-elect...
Reposted by Kyle
Apply to the new CSHL grad program in BioAI/neuroAI!

(Masters degree or equivalent required in a quantitative area such as computer science, physics, mathematics, or engineering)

www.cshl.edu/phd-program/...
Learning it aside, would you use it to submit a paper today?
Typst was honestly faster to learn up to my Latex proficiency. I think it’s cause there’s a clear programming model vs. latex packages are a mix of everything. It makes it so the intuitive thing is the correct thing.

Also: lilaq.org. Typst knowing about other file types is amazing.
Advanced data visualization in Typst − Lilaq
Advanced data visualization in Typst
lilaq.org
Reposted by Kyle
Moves as many people as every other transit agency in America put together! Twice as many people every day as the entire federal aviation system!
Reposted by Kyle
This is the biggest, toughest article I’ve done about so-called “AI psychosis.” It’s the story of a man who committed a horrific crime in his youth but served his time and against all odds found love and a new life—one that completely unraveled after he started talking to Google’s Gemini chatbot.
He Grew Obsessed With an AI Chatbot. Then He Vanished in the Ozarks
Jon Ganz committed a terrible crime in his youth, but he survived prison, fell in love, and started over. His new life unraveled in a way nobody could have predicted.
www.rollingstone.com
Reposted by Kyle
really neat clear explainer for the new on “centralizing flows” to theoretically model learning dynamics
Understanding Optimization in Deep Learning with Central Flows
centralflows.github.io
Reposted by Kyle
A lot of people think that every international student admitted means one fewer spot for domestic students, when the opposite is more likely true - the tuition revenue international students bring allows public universities to provide substantial discounts to domestic students, improving access.
Reposted by Kyle
🚨Our NeurIPS 2025 competition Mouse vs. AI is LIVE!

We combine a visual navigation task + large-scale mouse neural data to test what makes visual RL agents robust and brain-like.

Top teams: featured at NeurIPS + co-author our summary paper. Join the challenge!

Whitepaper: arxiv.org/abs/2509.14446
Mouse vs. AI: A Neuroethological Benchmark for Visual Robustness and Neural Alignment
Visual robustness under real-world conditions remains a critical bottleneck for modern reinforcement learning agents. In contrast, biological systems such as mice show remarkable resilience to environ...
arxiv.org
I mean the simple answer is that the team building the “100% agentic coding” product is going to attempt to reach that goal internally?
Reposted by Kyle
Do you plot transcriptomic data, coloring each cell by its cell type? ~ Don't use a default colormap! ~

Instead, use colormaps that capture biological meaning. If two cell types are very similar, their colors should be similar too. Read on 🧵

🧬💻
Reposted by Kyle
🚫TOMORROW'S FIGHT FOR SCIENCE RALLY POSTPONED🚫
We would like all our rallygoers to be safe and are postponing due to inclement weather!
Fret not — we will let you know the next opportunity to join us in the fight for health and science for the people!
Reposted by Kyle
Understanding the "why" of biology requires knowing a fair amount of what could rightly just be called history. Like, a specific thing happened a certain way at a certain time. This is not how people usually like to think about the study of biology...
And in both cases history tells you way more of the story than optimality
Tbh any large engineered system is the same. Evolutionary optimization dynamics will certainly add their own twist, but I think this is a property of any system with a large enough set of things it needs to do + some flexibility in how it could be done.
Reposted by Kyle
"evolutionary process is opportunistic, working with whatever variations arise, and it cannot be meaningfully equated with engineering toward an ideal."

It is funny how anyone who has worked with evolutionary algorithms knows they're not so great, yet we assume real evolution is a perfect optimizer
Reposted by Kyle
When reporting on these kinds of claims, the media should ask:
Reposted by Kyle
NEW: The anti-Trump protest happening right now on the streets of Washington DC is HUGE!

It looks like tens of thousands have joined.

This is the view from 16th Street - people stretching as far as the eye can see 👏
Reposted by Kyle
Lovely morning on the North Dublin Riviera! 😊
Reposted by Kyle
Results so stunningly clear they inspired this classic xkcd (xkcd.com/2400/):