Aatish Bhatia

Aatish Bhatia

@aatishb.com

I'm a science journalist who creates visually immersive interactive articles. Chasing the satisfaction of an aha moment. Previously: NYT, Princeton, Rutgers Physics Ph.D.

1.1K followers926 following317 posts

Aatish Bhatia reposted

For years, @aatishb.com and I have met weekly to talk (neuro)science. One question keeps coming back: what does it actually mean to find "the neural circuit for a behavior"? We've turned those conversations into a series that explains the success stories of neuroscience. Check out the first one👇🏽

Aatish Bhatia@aatishb.com

🧠✨ I have a new project out with my friend (and brilliant neuroscientist) @arkarupbanerjee.bsky.social exploring how collections of neurons create complex behaviors. In the first chapter, we wire together a brain from scratch to guide a creature towards the light. Check it out! aatishb.com/myriad/

Aatish Bhatia reposted

Three years ago, my dad said that all fish are the same shape. Out of spite, I decided to draw all 478 families of bony fish listed in Nelson 2016. There's still some stuff I need to edit and some fish I want to redraw but, after 985 days, I finished my big dumb fish project last night.

If you missed it: I wrote about why it takes such a massive amount of computation to train AI. The numbers are beyond astronomical.

Aatish Bhatia@aatishb.com

Have a new story out in NYT today about the mind-boggling amount of math calculations that it takes to train AI. Here's a paywall free link:

Aatish Bhatia reposted

Why do I struggle even reading papers that are clearly AI? Writing is a window into another person’s brain, their thinking, values, and outlook. When I read AI text, I mostly feel they shortchanged themselves and me by not offering that.

Aatish Bhatia reposted

Sorry to hear that Jeffrey Goldstone has died. He was a giant of quantum field theory- pioneering our modern understanding of symmetry breaking- and later did great things in quantum information. I knew him at MIT as a gruff but friendly wise man. physics.mit.edu/news/jeffrey...

physics.mit.eduJeffrey Goldstone, pioneering theoretical physicist and MIT professor emeritus, dies at 93The Physics Department is very sorry to report that our colleague Jeffrey Goldstone, Cecil and Ida Green Professor Emeritus of Physics, died on October 12 at the age of 93. Jeffrey began his career as...

Aatish Bhatia reposted

Aatish Bhatia reposted

Now on @sciam.bsky.social: OpenAI’s latest math dump has arrived. It consists of hundreds of new results to major problems, from the same internal model that produced a million-dollar math proof last month. It could take mathematicians years to sift through it all.

scientificamerican.comOpenAI unleashes hundreds more math results upon a field already in shockA month after resolving one of the six biggest open problems in mathematics, the company says its new internal model has toppled heaps more—this time without the multimillion-dollar price tagOpenAI un...

Aatish Bhatia reposted

Replying to @astrokatie.com

How rare are neutrino interactions? A neutrino (of the type emitted by the Sun) could go through a light year of lead with only a small chance of interacting. They pass through the Earth, and each of us, constantly, without anyone noticing. 100 trillion neutrinos go through you every second.

This is a really nice example of how LLMs learn shortcuts — heuristic strategies that can get the right answer for data within its training distribution. But as this example shows, these shortcuts can generalize poorly to data outside this distribution (e.g. here adding larger numbers).

Colin@colin-fraser.net

here's an example of a test I did over two years ago using gpt-4o. I am asking not only for it to compute the sum but to return the answer in words. 4o was not a so-called "thinking" model so there is no intermediate step between the request and the response. It just blurts out the sums.

I think the idea of shortcuts provides some helpful intuition beyond 'stochastic parrots'. The latter suggests that LLMs are merely recombining their training data. Shorcuts imply they are learning heuristic algorithms that generate such data (and these heuristics can fail in unexpected ways).

Aatish Bhatia reposted

This is a photograph of the edge of one of Saturn's rings, looking down from above. Those spiky dark shapes are shadows. Shadows of giant towers of ice and dust grains clumped together that rise up to three *kilometres* above the 10-m thick rings.

"If mathematics is genuinely about understanding rather than output, then this principle should appear not merely in declarations about AI but in hiring criteria, the evaluation process and journal practices. It is imperative to start changing the system now." proofsandprompts.com/2026/09/18/t...

proofsandprompts.comTwo responses to “A Severe Misalignment of AI in Mathematics”Note from P&P: On 11 September 2026, 25 Fields Medallists co-authored the declaration “A Severe Misalignment of AI in Mathematics”, which at the time of posting has over 7000 signat…

Aatish Bhatia reposted

Really interesting article by @aatishb.com on how LLMs work!

Aatish Bhatia@aatishb.com

NEW blog post about AI’s ‘jagged intelligence’: I trained an AI model from scratch to play a simple children’s puzzle. Unlike a human player, it can solve the hardest puzzles, but it fails to solve the easiest ones. aatishb.com/blog/2026/ai...

Aatish Bhatia reposted

As of today, the CDC has changed (reverted) their Covid guidance to make the vaccine recommended for all people over 18. In addition to being a smart public health move, this recommendation also means insurance must cover it. Read more here: www.cdc.gov/covid/hcp/va... Listen to a breakdown here:

Aatish Bhatia reposted

The Facial Expressions of Tethys by artist Julius von Bismarck. A suspended high-sea buoy that moves in real time. Sensors on a twin buoy out in the Atlantic Ocean send live satellite data to motors inside the museum. This makes the indoor buoy copy the exact, real-time movements of the ocean waves.

In other news, I started a new blog. Living my best internet life like it's 2015 :)

Aatish Bhatia@aatishb.com

NEW blog post about AI’s ‘jagged intelligence’: I trained an AI model from scratch to play a simple children’s puzzle. Unlike a human player, it can solve the hardest puzzles, but it fails to solve the easiest ones. aatishb.com/blog/2026/ai...

Aatish Bhatia reposted

#TodayinHistory #dataviz #OTD 📊 🎂Sep 22, 1791 Michael Faraday born in Newington Butts, Surrey, England 🏴󠁧󠁢󠁥󠁮󠁧󠁿 1820-1835: Graphical analysis applied to natural phenomena (electric and magnetic lines of force, weather, tides) Made these invisible constructs open to inspection 👓

Aatish Bhatia reposted

So chilling - but thanks @elisecutts.bsky.social for the roundup and thoughtful commentary. "If AI scams — and even just lazy AI pitches that flood inboxes — force us to rely on relationship networks to establish trust, science journalism is going to close up." www.reviewertoo.com/2026-09-21-s...

reviewertoo.comA scammer impersonated a science journalist. It worked.AI fraud is going to make it harder for new voices to break into science journalism

Love this idea, a great way to visualize how far we scroll

shri@shrikhalpada.dev

IMO the absolute bare minimum of social media regulation should involve platforms telling you exactly how long you've been scrolling, e.g. "you are 96 videos into your feed" or, using the length of a CVS receipt as a proxy! this was the idea behind a thought experiment I made a few months ago

Aatish Bhatia reposted

LLMs are going to fundamentally change what scholarship is. Pumping out true propositions used to be a viable career. Now much of that can be offloaded, leaving scholars the tasks of understanding, synthesizing, and deciding what is valuable. Goes for math, philosophy, parts of physics.

This post is unavailable.

Thoughtful piece: "Upon reflection, I've come to believe that quality software is not about how many features something has or how quickly it is built but rather how cared for it is by the designers, developers, and product people who work on it."

"There's an ineffable aspect to my favorite pieces of software where I don't think of it is as feature matrix but rather as an interface made by a human for humans that's a joy to use. And when you build software with AI, it leads to worse software." brettcodes.com/im-done-usin...

Aatish Bhatia reposted

To tell if a maze is solvable, just hang it by its corners If it tears into pieces, you’ve found a solution