Martin Wattenberg
@wattenberg.bsky.social
Human/AI interaction. ML interpretability. Visualization as design, science, art. Professor at Harvard, and part-time at Google DeepMind.
Reposted by Martin Wattenberg
Charts and graphs help people analyze data, but can they also help AI?
In a new paper, we provide initial evidence that it does! GPT 4.1 and Claude 3.5 describe three synthetic datasets more precisely and accurately when raw data is accompanied by a scatter plot. Read more in🧵!
In a new paper, we provide initial evidence that it does! GPT 4.1 and Claude 3.5 describe three synthetic datasets more precisely and accurately when raw data is accompanied by a scatter plot. Read more in🧵!
August 4, 2025 at 2:17 PM
Charts and graphs help people analyze data, but can they also help AI?
In a new paper, we provide initial evidence that it does! GPT 4.1 and Claude 3.5 describe three synthetic datasets more precisely and accurately when raw data is accompanied by a scatter plot. Read more in🧵!
In a new paper, we provide initial evidence that it does! GPT 4.1 and Claude 3.5 describe three synthetic datasets more precisely and accurately when raw data is accompanied by a scatter plot. Read more in🧵!
Reposted by Martin Wattenberg
AI is often thought of as a black box -- no way to know what's going on inside. That's changing in eye-opening ways. Researchers are finding "beliefs" models are forming as they converse, and how those beliefs correlate to what the models say and how they say it.
www.theatlantic.com/technology/a...
www.theatlantic.com/technology/a...
What AI Thinks It Knows About You
What happens when people can see what assumptions a large language model is making about them?
www.theatlantic.com
May 21, 2025 at 3:32 PM
AI is often thought of as a black box -- no way to know what's going on inside. That's changing in eye-opening ways. Researchers are finding "beliefs" models are forming as they converse, and how those beliefs correlate to what the models say and how they say it.
www.theatlantic.com/technology/a...
www.theatlantic.com/technology/a...
Reposted by Martin Wattenberg
The interactive NameGrapher is updated with 2024 baby name popularity stats! Come explore--and marvel that Oliver and Olivia have converged namerology.com/baby-name-gr...
May 12, 2025 at 7:09 PM
The interactive NameGrapher is updated with 2024 baby name popularity stats! Come explore--and marvel that Oliver and Olivia have converged namerology.com/baby-name-gr...
A wonderful visualization for those of us obsessed by sunlight and geography!
This map shows the hour of sunrise globally through the year. It reveals time zones following national and, sometimes, regional boundaries, and slicing through the oceans.
May 12, 2025 at 1:36 PM
A wonderful visualization for those of us obsessed by sunlight and geography!
An incredibly rich, detailed view of neural net internals! There are so many insights in these papers. And the visualizations of "addition circuit" features are just plain cool!
Can we understand the mechanisms of a frontier AI model?
📝 Blog post: www.anthropic.com/research/tra...
🧪 "Biology" paper: transformer-circuits.pub/2025/attribu...
⚙️ Methods paper: transformer-circuits.pub/2025/attribu...
Featuring basic multi-step reasoning, planning, introspection and more!
📝 Blog post: www.anthropic.com/research/tra...
🧪 "Biology" paper: transformer-circuits.pub/2025/attribu...
⚙️ Methods paper: transformer-circuits.pub/2025/attribu...
Featuring basic multi-step reasoning, planning, introspection and more!
On the Biology of a Large Language Model
transformer-circuits.pub
March 27, 2025 at 8:00 PM
An incredibly rich, detailed view of neural net internals! There are so many insights in these papers. And the visualizations of "addition circuit" features are just plain cool!
Now that we have vibe coding, we need vibe testing!
March 24, 2025 at 7:45 PM
Now that we have vibe coding, we need vibe testing!
The wind map at hint.fm/wind/ has been running since 2012, relying on weather data from NOAA. We added a notice like this today. Thanks to @cambecc.bsky.social for the inspiration.
March 3, 2025 at 1:56 AM
The wind map at hint.fm/wind/ has been running since 2012, relying on weather data from NOAA. We added a notice like this today. Thanks to @cambecc.bsky.social for the inspiration.
Neat visualization that came up in the ARBOR project: this shows DeepSeek "thinking" about a question, and color is the probability that, if it exited thinking, it would give the right answer. (Here yellow means correct.)
February 25, 2025 at 6:44 PM
Neat visualization that came up in the ARBOR project: this shows DeepSeek "thinking" about a question, and color is the probability that, if it exited thinking, it would give the right answer. (Here yellow means correct.)
Great thread describing the new ARBOR open interpretability project, which has some fascinating projects already. Take a look!
ARBOR aims to accelerate the internal investigation of the new class of AI "reasoning" models.
See the ARBOR discussion board for a thread for each project underway.
github.com/ArborProjec...
See the ARBOR discussion board for a thread for each project underway.
github.com/ArborProjec...
February 20, 2025 at 10:49 PM
Great thread describing the new ARBOR open interpretability project, which has some fascinating projects already. Take a look!
Today we're launching a multi-lab open collaboration, the ARBOR project, to accelerate AI interpretability research for reasoning models. Please join us!
github.com/ARBORproject...
(ARBOR = Analysis of Reasoning Behavior through Open Research)
github.com/ARBORproject...
(ARBOR = Analysis of Reasoning Behavior through Open Research)
GitHub - ARBORproject/arborproject.github.io
Contribute to ARBORproject/arborproject.github.io development by creating an account on GitHub.
github.com
February 20, 2025 at 7:55 PM
Today we're launching a multi-lab open collaboration, the ARBOR project, to accelerate AI interpretability research for reasoning models. Please join us!
github.com/ARBORproject...
(ARBOR = Analysis of Reasoning Behavior through Open Research)
github.com/ARBORproject...
(ARBOR = Analysis of Reasoning Behavior through Open Research)
Out: rectangular maps of the globe
In: rectangular maps of butterfly wings!
In: rectangular maps of butterfly wings!
In 1897, Alfred G. Mayer created his butterfly wing projections, an attempt to gain new insights into natural patterns and laws. Vertical blocks denote individual wings, distorted and stretched mathematically to fill a tidy rectangular space. More here: publicdomainreview.org/collection/m...
January 19, 2025 at 7:48 PM
Out: rectangular maps of the globe
In: rectangular maps of butterfly wings!
In: rectangular maps of butterfly wings!
I'm realizing that many people hear "generative art" (a term going back at least to the 60s!) as synonymous with generative AI. Is this how cryptographers feel about the new meaning of "crypto"?
January 17, 2025 at 12:35 PM
I'm realizing that many people hear "generative art" (a term going back at least to the 60s!) as synonymous with generative AI. Is this how cryptographers feel about the new meaning of "crypto"?
Reposted by Martin Wattenberg
What a beauty! This is comet C/2024 G3 (ATLAS) passing through the field of view of the LASCO C3 coronagraph.
It wasn't for certain whether it would survive it's closest approach to the sun on January 13th, but it did and delivered us a spectacular show!
#comet #C2024G3 🔭
It wasn't for certain whether it would survive it's closest approach to the sun on January 13th, but it did and delivered us a spectacular show!
#comet #C2024G3 🔭
January 16, 2025 at 9:14 PM
Idle question: Are there any papers from legit math journals using emoji as notation?
We already use every symbol on the keyboard, musical sharps and flats, and even weird made-up fonts (what is that Weierstrass P??). A smiley is easy to draw with chalk and put into LaTeX, so why not?
We already use every symbol on the keyboard, musical sharps and flats, and even weird made-up fonts (what is that Weierstrass P??). A smiley is easy to draw with chalk and put into LaTeX, so why not?
January 15, 2025 at 2:03 PM
Idle question: Are there any papers from legit math journals using emoji as notation?
We already use every symbol on the keyboard, musical sharps and flats, and even weird made-up fonts (what is that Weierstrass P??). A smiley is easy to draw with chalk and put into LaTeX, so why not?
We already use every symbol on the keyboard, musical sharps and flats, and even weird made-up fonts (what is that Weierstrass P??). A smiley is easy to draw with chalk and put into LaTeX, so why not?
I tried asking AI for writing advice and it was so sarcastic I'm never going to ask again
January 5, 2025 at 6:42 PM
I tried asking AI for writing advice and it was so sarcastic I'm never going to ask again
Reposted by Martin Wattenberg
New paper <3
Interested in inference-time scaling? In-context Learning? Mech Interp?
LMs can solve novel in-context tasks, with sufficient examples (longer contexts). Why? Bc they dynamically form *in-context representations*!
1/N
Interested in inference-time scaling? In-context Learning? Mech Interp?
LMs can solve novel in-context tasks, with sufficient examples (longer contexts). Why? Bc they dynamically form *in-context representations*!
1/N
January 5, 2025 at 3:49 PM
New paper <3
Interested in inference-time scaling? In-context Learning? Mech Interp?
LMs can solve novel in-context tasks, with sufficient examples (longer contexts). Why? Bc they dynamically form *in-context representations*!
1/N
Interested in inference-time scaling? In-context Learning? Mech Interp?
LMs can solve novel in-context tasks, with sufficient examples (longer contexts). Why? Bc they dynamically form *in-context representations*!
1/N
If you're the kind of person who likes thinking about the sky (and I am!) this is a wonderful visualization to explore. So many patterns, each with its own story.
For the 4th year in a row, my all-sky camera has been taking an image of the sky above the Netherlands every 15 seconds. Combining these images reveal the length of the night changing throughout the year, the passage of clouds and the motion of the Moon and the Sun through the sky. #astrophotography
January 5, 2025 at 1:28 PM
If you're the kind of person who likes thinking about the sky (and I am!) this is a wonderful visualization to explore. So many patterns, each with its own story.
In art and literature, "criticism" doesn't mean "pointing out flaws." It's something bigger and more interesting than a referee calling fouls. I think we should have the same ambitions for data visualization criticism!
January 3, 2025 at 7:32 PM
In art and literature, "criticism" doesn't mean "pointing out flaws." It's something bigger and more interesting than a referee calling fouls. I think we should have the same ambitions for data visualization criticism!
Reading this 1954 court case, and am wondering if English professors are ever called as expert witnesses to testify whether a character is a mere "chessman" in a story
scholar.google.com/scholar_case...
scholar.google.com/scholar_case...
January 3, 2025 at 3:51 PM
Reading this 1954 court case, and am wondering if English professors are ever called as expert witnesses to testify whether a character is a mere "chessman" in a story
scholar.google.com/scholar_case...
scholar.google.com/scholar_case...
Made good progress on my to-do list, with the exception of "One (1) new idea"
December 30, 2024 at 12:29 AM
Made good progress on my to-do list, with the exception of "One (1) new idea"
A book recommendation! I just finished "Get the Picture" by Bianca Bosker, a dive into the extremely peculiar world of contemporary art. Amazingly well-observed, well-written, and just plain fun to read.
www.penguinrandomhouse.com/books/602064...
www.penguinrandomhouse.com/books/602064...
Get the Picture by Bianca Bosker: 9780525562207 | PenguinRandomHouse.com: Books
AN INSTANT NEW YORK TIMES BESTSELLER NAMED A BEST BOOK OF 2024 BY NPR, TIME, AND THE ECONOMIST “Get the Picture is one of the funniest books I’ve read . . . Brilliant.” —The...
www.penguinrandomhouse.com
December 27, 2024 at 9:16 PM
A book recommendation! I just finished "Get the Picture" by Bianca Bosker, a dive into the extremely peculiar world of contemporary art. Amazingly well-observed, well-written, and just plain fun to read.
www.penguinrandomhouse.com/books/602064...
www.penguinrandomhouse.com/books/602064...
The math benchmarks I want:
1. OopsBench: given a faulty proof with numbered steps, which step contains an unfixable logical flaw?
2. DunnoMath: half the problems are taken from FrontierMath, half are almost certainly unsolvable. Major points off for guessing an answer to an unsolvable problem.
1. OopsBench: given a faulty proof with numbered steps, which step contains an unfixable logical flaw?
2. DunnoMath: half the problems are taken from FrontierMath, half are almost certainly unsolvable. Major points off for guessing an answer to an unsolvable problem.
December 23, 2024 at 2:21 PM
The math benchmarks I want:
1. OopsBench: given a faulty proof with numbered steps, which step contains an unfixable logical flaw?
2. DunnoMath: half the problems are taken from FrontierMath, half are almost certainly unsolvable. Major points off for guessing an answer to an unsolvable problem.
1. OopsBench: given a faulty proof with numbered steps, which step contains an unfixable logical flaw?
2. DunnoMath: half the problems are taken from FrontierMath, half are almost certainly unsolvable. Major points off for guessing an answer to an unsolvable problem.
Fractions can be weirdly beautiful, for something so mundane. This visualization just plots points of the form (a/b, c/d). Bigger dots mean smaller denominators. The biggest dot is (0, 0).
December 20, 2024 at 3:13 AM
Fractions can be weirdly beautiful, for something so mundane. This visualization just plots points of the form (a/b, c/d). Bigger dots mean smaller denominators. The biggest dot is (0, 0).
You can fry ChatGPT's circuits by asking a question in Morse code and telling it to answer only in Morse code. Yet Claude doesn't even blink. Huh. The question: "Which character in the movie Groundhog Day do you identify with the most, and why?" First translated result is ChatGPT; second is Claude.
December 19, 2024 at 5:47 PM
You can fry ChatGPT's circuits by asking a question in Morse code and telling it to answer only in Morse code. Yet Claude doesn't even blink. Huh. The question: "Which character in the movie Groundhog Day do you identify with the most, and why?" First translated result is ChatGPT; second is Claude.
Back in 2009, a site called Wordle let you make word clouds. Paste in some writing, choose a few options, and bingo: a beautiful tessellation of vertical and horizontal text. We did a survey of its users, and the headline result was that the vast majority of people using it felt "creative."
December 18, 2024 at 3:57 AM
Back in 2009, a site called Wordle let you make word clouds. Paste in some writing, choose a few options, and bingo: a beautiful tessellation of vertical and horizontal text. We did a survey of its users, and the headline result was that the vast majority of people using it felt "creative."