Jonathan Cheng

A ton of attention over the years goes to plots comparing open to closed models.
The real trend that matters for AI impacts on society is the gap between closed frontier models and local consumer models.
Local models passing major milestones will have major repercussions.
buff.ly/ccMJydQ

Dark blue = most
Gray = middle
Light blue = least?!?

Please.

Someone get ucla law on the line, cause this data viz’s color choice is hot garbo.

If it provides any comfort/ assuages imposter syndrome, I just learned about a very smart person who wrote pedabytes of NaNs, bc of a bug in their pipeline.

Assistant Professor - Information - School of Information University of California, Berkeley is hiring. Apply now!

The UC Berkeley School of Information is hiring an assistant professor in the broad field of Information--including areas of info seeking/retrieval, digital humanities, cultural analytics, info viz, & philosophy of information (among others). Deadline Nov 1! aprecruit.berkeley.edu/JPF05014

Also, if you haven’t imagined being haunted in one of these buildings, you haven’t truly lived in the Midwest.

They were selling these in the gift shop in the Seattle art museum.

And, my god, the restraint I showed? Applaudable. *waits for applause*

But also, these are absolutely stunning.

“You get 100 flops per minute!”
“Floating point operations?”
“What? Pfft. No. Fish- lopped per minute!”

You might think tpus/GPUs are complicated, but have you tried operating a 1910 “fish processing machine”?

The animals at the Seattle Zoo are *sleepy*

Big warning for all weebs out there. This is a code red.

Somewhere, my Korean and Chinese ancestors are like “oh yeah? You like ordering Japanese things? GET TARRIFFED, SON!”

Comparing AI Labs in the United States and China, by Jingyuan Liu

🔹Multi-Token Prediction → turbo-charged speculative decoding

Demo: chat.qwen.ai
Blog: qwen.ai/blog?id=4074...
Model - Huggingface: huggingface.co/collections/...
Model - Kaggle: www.kaggle.com/models/qwen-...
Alibaba Cloud API: www.alibabacloud.com/help/en/mode...

I’ve lost power in Seattle for like the third time in a month.

Which is not helping my nostalgia for Chicago’s better transit & stable power grid 😂

Otoh, there are some pretty church buildings in the area.

YouTube video by Jane Street Building Machine Learning Systems for a Trillion Trillion Floating Point Operations

Super fun high-level talk on the history of ml frameworks.

18:50 is a nice overview of torch compilers
19:10: has a nice demarcation of user/execution categories
21:30 nicely describes, why GPUs?

Nicely goes between broader historical effects, software, and hardware.

youtu.be/139UPjoq7Kw?...

I really enjoy working from cafes, and zero trust network access/ ip reputation filtering/ geofencing is going to ruin me 😂

There are some perks of my job switch…but having just visited the Seattle Riot office — work somewhere with a dope IP, the work spaces are so great

At long last, I can post my team's summer project: applied modules to teach how ML/AI tools are changing social science and humanities research: ubcecon.github.io/praxis-ubc/

Highlights:

LegalBERT to analyze anti/pro-immigrant sentiment in 19th c. BC law: ubcecon.github.io/praxis-ubc/d...
🧵1

My slogan to young DHers would be “vibe code till dawn, there’s intellectual ground for the taking!”

+1 I just think that a field that has both coded and taught young readers/writers for years — thrives in an era when ml papers fumble about their “rephrasing” strategies and preference datasets have not been treated with some basic stylometry.

In hindsight, what I should’ve done is written a few articles or a book, *and then* joined an outfit where I shouldn’t say much. There’s *a lot* to say here.

Something something, one person’s language data augmentation is another person’s distant writing is another person’s creative writing project

Yeah, I know exactly what you mean. Given that *nervous cough* I suspect we’re thinking about very similar problems.

There’s a line to draw from before nanogenmo, back translation, synthetic data making, etc etc.

It’s both interesting and, frankly, very fun.

Seattle, advertising itself to be the setting of many fog-related thrillers.

I thought I ordered Chinese food…but I might be receiving way more than I bargained for….

Just made a playlist called “The Goats of Melancholy,” and I really wish I had the musical talent needed to build a band on that name.

Now look, I can understand, just because transparent does not mean weak.

But my primate brain, otoh, will not step on this.

GitHub - google-deepmind/limit: On the Theoretical Limitations of Embedding-Based Retrieval On the Theoretical Limitations of Embedding-Based Retrieval - google-deepmind/limit

"On the Theoretical Limitations of Embedding-based Retrieval"

Paper: arxiv.org/abs/2508.21038
Repo: github.com/google-deepm...

💯 the artists/ designers at riot games should really take pride that they were heavily influential in this