Joined November 2010
18 Photos and videos
Pinned Tweet
Ever wanted to generate smokey logograms from the heptapod language in the awesome sci-fi movie "Arrival" (2016)? Probably not 😅 But a few people asked for the text-to-image model I trained, so it's now on @replicate ! replicate.com/meghabyte/arri…
15
41
313
104,952
Megha Srivastava retweeted
By popular demand, Claude Code-written JavaScript Warli generator.
Claude Code wrote a JavaScript Madhubani generator.
18
42
415
21,265
Megha Srivastava retweeted
Sad to hear that Ada's Technical Books in Seattle is closing this week -- it's a wonderful bookstore. I always found something obscure but interesting when I visited. adasbooks.com

33
13
399
69,414
Megha Srivastava retweeted
i mounted a tiny microphone on my apartment balcony to listen for any birds passing by and built a site to collage them as they're heard
271
868
14,300
1,883,344
Megha Srivastava retweeted
made a site that picks the closest rothko for how the weather feels outside your window
172
1,480
15,371
1,368,448
Back in Seattle at #MLSys2026 to present Hawkeye, a tool for exact replication of GPU-level operations (e.g. FP16 matmul on NVIDA Hopper) on a CPU, useful for verifiable ML! 🔐 w/ Erez Badash, Dan Boneh, @komargodski Paper: arxiv.org/abs/2603.20421 Code: github.com/badasherez/gpu-si…
13
59
17,018
My Data Sonification project mapping hourly forest data (e.g. soil moisture, temp) to the bansuri 🪈 received 1st place at @HubbardBrookNH 's data contest! 🎶🌲 It was fun figuring out how to make real-world climate data "sound" good 🙂Check it out here! meghabyte.github.io/blog/ban…
2
8
37
9,231
Megha Srivastava retweeted
i met the woman who takes care of the wild parrots in sf her home is filled with photos of them, close up shots, their heads tilted to the side, looking directly at the camera. she knows their routines, where they are in the city depending on the time of day. they all have names. she knows who is partnered with who. they bring her their children in the summer, when they are nesting. she feed them apples and raw sunflower seeds and walnuts. i watched one eat an apple out of her hand. his name is romeo. he brought his son with him. apparently romeo is looking for his fourth wife now.
24
147
2,318
47,326
Megha Srivastava retweeted
Eaton Fire burn scar 16 months later. The fire blasted it's way through here at 60 MPH with temperatures of 1,500°F - 2,000°F. California native plants rock.
27
311
3,614
65,025
Megha Srivastava retweeted
March cherry blossom Madness. Tacoma, Washington
18
433
4,576
86,605
Check out our paper taking a broader look at emergent misalignment in large language models, published today in @Nature !
Replying to @OwainEvans_UK
Today we publish an extended paper on emergent misalignment in Nature. We communicate our original insecure-code results to a broad audience and incorporate the findings from many other researchers. It's open access! nature.com/articles/s41586-0…
2
2
15
2,410
Megha Srivastava retweeted
9 Nov 2025
Blue Cross had refused to pay thousands of claims for the center’s patients, but on several occasions execs at the insurance company had signed special one-time deals with the center to pay for their wives’ cancer treatment. propublica.org/article/blue-…
17
694
975
53,212
I'll be discussing work on using AI for pharmacovigilance (led by @claudeai 😉 ) as a Spotlight in #Agents4Science today at 11:00am PST! It was really fun participating in @james_y_zou's experiment on what happens if we let an LM lead a research project agents4science.stanford.edu
3
11
54
10,527
Megha Srivastava retweeted
I can't stop thinking about Adrienne Rich's poem about AI, written in 1961.
23
282
1,620
61,327
New preprint on AI Education! 🍎 “Modeling Student Learning with 3.8M Program Traces” 💻 When students code, their edits tell a story about their reasoning process: exploring, debugging, and tinkering 🧠 What can LMs learn from training on student edit sequences? 📚
9
32
210
55,855
💡 Takeaway: Many properties of code are properties of the student coder. LMs trained on students’ edit traces are more generalizable, controllable, and predictive of diverse student behavior. Paper: arxiv.org/abs/2510.05056 Joint work w/ @alexisjross , @DoctorJei, @jacobandreas

1
1
7
873