Bing Yan

AI+Chemistry PhD student @nyuniversity | Prev PhD of Chemistry @mit

[Kyunghyun Cho @kchonyc]

seriously NYU's visitor registration portal based on JRNY sucksssssssssssss! i wasted so much time working with it every week.

so ... here's a more modern, user friendlier version for you: install it on your browser with tampermonkey today. the link below.

you're welcome, @nyuniversity !

[Bing Yan @bingyan4science]

Electron Flow Matching for Generative Reaction Mechanism Prediction (2025)
https://t.co/4Jgn5nA5E9

[Bing Yan @bingyan4science]

Great Papers, Great Framing.

#3 Electron Flow Matching for Generative Reaction Mechanism Prediction (2025)

1/ What is the frame?

Mass conservation is a fundamental principle, is often violated in ML models.

By modeling electron redistribution, FlowER explicitly satisfies it.

[Bing Yan @bingyan4science]

4/ What can we learn?

In AI4Science, it is easy to fall into an incremental frame:

"We applied method X to domain Y."

A much more memorable frame shows insights behind the design:

"What principle does the current model miss?"
"We build the model around that principle."

[Bing Yan @bingyan4science]

Language Models are Few-Shot Learners
https://t.co/CVyPI4CIXM

[Bing Yan @bingyan4science]

Great Papers, Great Framing.

#2 Language Models are Few-Shot Learners (2020)

(GPT-3)

1/ What is the frame?

The frame is not:

"We trained a bigger language model."

The frame is:

LLMs can adapt to new tasks through text interaction alone, without task-specific fine-tuning.

[Bing Yan @bingyan4science]

4/ What can we learn?

Great framing is not the most extreme claim.

It is the one that best captures the mechanism.

For GPT-3, the mechanism is:

"as models scale, they are better at using context as task specification."

The story is not size.

The story is what size unlocks.

[Yuntian Deng @yuntiandeng]

https://t.co/vgGYhGpHhx

[Bing Yan @bingyan4science]

Attention Is All You Need.
https://t.co/TnV33QSL6z

[Bing Yan @bingyan4science]

I'm starting a new series:

Great Papers, Great Framing.

I'll analyze 1 paper daily and ask 4 questions:

-What is the frame?
-Why is it powerful?
-What is a weaker frame?
-What can we learn?

The goal is to learn how to tell a scientific story.

#1 Attention Is All You Need

[Bing Yan @bingyan4science]

4/ What can we learn?

Many great papers don't start with a better solution.

They start by identifying an assumption everyone accepts.

Then they ask:

"What if that assumption is wrong?"

[Yuntian Deng @yuntiandeng]

I built a 3D character you can control with language instead of predefined buttons.

How? I compiled a neural program that turns language instructions into movements, using ProgramAsWeights.

Just type: "act excited, wave, dance, then sit proudly"

Try it: https://t.co/wnQcwbDJ3M

[Zack Ulissi @zackulissi]

UMA playground for interactive chemistry!

https://t.co/CLBcLYLdQW

Get in touch if you have fun ideas for how to use this for education.

And if you have a favorite undergraduate or high school chemistry teacher who might find this useful, please share with them!