Researcher || Lecturer || DL Theory, GNN, Human Action Recognition, Network Science || British Humor || Rowing || Humanist
Sri Lanka
Joined November 2013
- Tweets2.3K
- Following2.1K
- Followers155
- Likes4.4K
Sanka Mohottala retweeted
deep-learning-alchemy.github…
Such a cool way to teach ML in the age of AI.
- Assignments: run experiments, worth 0% of your grade, just to train intuitions.
- Quizzes: predict new experiment outcomes.
> Isnt it unfair to predict arbitrary experiments? Why do i care if the loss is 3.1234…
> We will pick outcome metrics that are relevant ... But also, the real world is unfair: your experiments depend on your prediction
Sanka Mohottala retweeted
🇱🇰🚁 Eight months in captivity: Sri Lanka Air Force helps rescue three hostages in Africa
The Sri Lanka Air Force has helped secure the release of three civilians held hostage for over eight months in the Central African Republic.
The high-risk UN peacekeeping operation involved three Mi-17 helicopters, combat personnel, a specialised negotiation team and medical teams.
Sri Lankan Air Force helicopters carried out more than 18 hours of flying operations, helping bring all three hostages to safety.
Sanka Mohottala retweeted
"Theories of Deep Learning" is a 2026–27 mathematics course offered by the University of Oxford that examines the mathematical foundations of deep neural networks.
The course consists of 16 freely available lectures covering neural network architectures, nonlinear transformations, convolutional structures, pooling, optimization, dictionary learning, transferability of learned features, adversarial examples, random matrix theory, learning invariances, and more.
The lectures explore mathematical theories that help explain how deep neural networks learn and why they perform as they do. I consider this a particularly valuable resource for anyone interested in the mathematics underlying modern AI systems.
courses.maths.ox.ac.uk/cours…
Sanka Mohottala retweeted
I can't sleep after this shit... Claude Code just found 12,154 photos of my face on the old Android I used for 8 years. I never took a single one of them.
The front camera was firing almost every time I picked the phone up. In bed, on the train, in the car, or when I just grabbed it to check the time.
They were real camera frames sitting in a hidden folder, each one with a timestamp and a location. None of them were in the Gallery or in Recently Deleted.
I plugged the phone into my laptop, turned on ADB and told Claude Code to look for front-camera files that weren't in my photos. It pulled a folder inside Android/data that belonged to a flashlight app I installed in 2019 and completely forgot about. 12,154 files.
I started scrolling. Almost all of it was from the last 15 months. Then I saw a file created 2 seconds ago, while I was still looking at the folder. The phone kept taking photos of me as I was reading this.
On Android 8 any app with camera permission can shoot in the background, and you get no green dot or any other warning when it does. That flashlight app had camera and location access from day one.
I uninstalled it, took camera access away from everything that doesn't need it and cleared the folder. How many of these photos already went somewhere else, I don't know.
If you want to check yours, plug it in, turn on USB debugging and ask Claude Code to find front-camera files outside your Gallery and show which apps used the camera and when.
Sanka Mohottala retweeted
One of my greatest intellectual inspiration is Samuel Eilenberg, co-creator of category theory, known for his work in algebraic topology and also, for his incredible art collection!
He also happened to have written one of the greatest book on theoretical computer science, which I am recommending today, titled ''Automata, Languages and Machines Volume A/B''.
This book was, in some way, quite visionary, unsurprising from Eilenberg who's mind consistently looked further than most before him.
Don't worry about the boring cover, if you know some abstract algebra, logic and theory of computation, you're going to have fun.
Sanka Mohottala retweeted
Graph Machine Learning course
Lecture 10 presents GNNs Meet Large Language Models 🌟
Text-attributed graphs, open-weight LLM reasoning, LLM-to-LM interpreter, LoRA adapter, GraphRAG, graph soft-token, G-Retriever.
storage.googleapis.com/xavie…
Sanka Mohottala retweeted
This is arguably one of the best technical books you can read on how LLMs are trained at scale:
> GPU memory and profiling
> tiling, kernel fusion, FlashAttention
> Data / Tensor / Pipeline / Context Parallelism
You can read the PDF here:
Sanka Mohottala retweeted
600km. Across Sri Lanka. On foot. 🇱🇰
British vet Dr Janey Lowes is taking on one seriously bonkers challenge, running the entire length of Sri Lanka to raise money for a new veterinary hospital for street dogs! 🐾
Through the heat, rain and mountains, she’s doing it all for the animals she’s dedicated her life to helping.
We’ve previously worked with and raised money for @WECare_SriLanka during our tours to the country, so this is a cause that already means a lot to us ❤️
An incredible effort for an incredible cause. Show Janey some support, if you can! 👏
Full details here ➡️ tinyurl.com/ycym6ym4
Sanka Mohottala retweeted
Algorithms by Jeff Erickson - pair it along with this MIT's adv data structures.
The illustrations are simply great, and this is one of the best algorithm books out there.
jeffe.cs.illinois.edu/teachi…
Sanka Mohottala retweeted
Hi all, I have a new result I'm excited to share. It turns out that *all existing implementations* of softmax and cross-entropy (that's JAX, PyTorch, textbooks, etc) have a subtle bug --- two bugs, actually. A simple rewrite fixes both, and eliminates "slingshots" when grokking:
Sanka Mohottala retweeted
To bring quantum applications within reach sooner, we must compress the circuits that power them. Join our @Kaggle competition to build efficient, verifiable circuits using our new open source kickmix library: goo.gle/4BqF2QX
ALT Graphic reading "Verifiable Quantum Circuit Compilation Competition with Google Quantum AI and Kaggle" featuring a close-up image of a leaf. At the bottom, text reads "Enter the competition on Kaggle."
Sanka Mohottala retweeted
University of Toronto's "Neural Networks for Machine Learning"
by Geoffrey Hinton
Videos: cs.toronto.edu/~hinton/cours…
Slides: cs.toronto.edu/~hinton/cours…
Sanka Mohottala retweeted
全世界包括清北的编程课程都应该向斯坦福的CS146S Fall 2026看齐！这门课相比一年前的Fall 2025已经砍掉了85%的内容，可以知道编程领域进展有多恐怖了，如果还停留在教科书十年不需更新的幻觉中，可想而知会多落后了！
两版课纲我逐周对了一遍：
第 1 周：教你怎么写 prompt → 200 行代码手搓一个 Claude Code
期末项目占分：80% → 50%
给真实开源项目提 PR：0 → 30%
去年还单独讲一周的 AI IDE、Warp 终端、一句 prompt 生成整个应用，今年全部都拿掉。
取而代之的是 Agent Skills、CLAUDE.md / AGENTS.md / hooks、agent-ready 代码库、后台并行 agent、MCP portal 和模型路由，最后一周单独讲软件工厂。
刚入门的话，我建议从第 1 周跟起：
9/22 那节 200 行手写 coding agent，slides 和完整代码都公开了，最核心的是 agent 原型—— read / write / edit / bash 四个工具加一个循环
9/24 拆主流 coding agent 的 system prompt，原始 prompt 文件也已放出来
如果已经天天跟 coding agent 打交道，前几周大概率扫一眼就行，我会重点盯这几块：
- 嘉宾：SpaceX SI负责人 @leerob、Claude Code之父 @bcherny 已经分享完，10/22 轮到 Factory 的 CTO @EnoReyes，后面还有 Cognition、Semgrep、Cloudflare、Replit 等的 @amasad 和 @eladgil
- 第 9–10 周：大团队怎么落地，MCP portal、LLM gateway、模型路由、成本优化
- 前五周的阅读清单，凑起来就是一份 2026 版 agent 工程书单
30% 的开源分也很有意思：Vercel、Semgrep、OpenHands、Unsloth 等 15 个开源项目给学生做指导，PR 直接往生产级代码库里提，这在大学课程里还是头一回。
课程材料陆续公开，不是斯坦福学生也能跟：themodernsoftware.dev
后面几周的嘉宾分享我会边看边拆，关注我，第一时间同步最佳实践🫡
I’m excited to finally announce the newest edition my Stanford course 𝗧𝗵𝗲 𝗠𝗼𝗱𝗲𝗿𝗻 𝗦𝗼𝗳𝘁𝘄𝗮𝗿𝗲 𝗗𝗲𝘃𝗲𝗹𝗼𝗽𝗲𝗿. It has been 9 months in the making.
Last November, with the release of Claude Opus 4.5, coding agents experienced a step function improvement in capability. We all felt it. The LLMs were more powerful, could reason for longer, solve harder tasks.
This year’s iteration of my course reflects the 2026 metamorphosis of software engineering.
My core belief is simple: AI-native developers of the LLM era are going to become the most important members of any software organization. I have designed my course to train this next generation of engineers.
𝗪𝗵𝗮𝘁’𝘀 𝗱𝗶𝗳𝗳𝗲𝗿𝗲𝗻𝘁 𝘁𝗵𝗶𝘀 𝘁𝗶𝗺𝗲 𝗮𝗿𝗼𝘂𝗻𝗱
First, 85% of my Fall 2025 class material is being thrown out. The Fall 2026 syllabus reflects the core capabilities AI-native engineers must have: agent skills, advanced context engineering, MCP portals, agent-ready codebase principles, agentic code review, security, parallelizing background agents, software factories, and more.
Second, I am going to teach my students how to have software taste. Every student will be required to ship pull requests to production-grade, real-world codebases. The course is collaborating with the top open-source AI repos who will offer support and mentorship to students on how to meaningfully contribute to their projects.
This has never been done before in any university course so I am incredibly grateful to our OSS Partners: @browserbase, @HeyGen, @CopilotKit, @semgrep, @OpenHandsDev, @milvusio, @marimo_io, Pi, @crewAIInc, @warpdotdev, @vercel, @cmux, @arizeai, @UnslothAI, and @anyscalecompute.
𝗪𝗵𝗮𝘁’𝘀 𝘀𝘁𝗮𝘆𝗶𝗻𝗴 𝘁𝗵𝗲 𝘀𝗮𝗺𝗲
I’m fortunate to again have AI software engineering leaders and founders as guest speakers to share their learnings from building top coding agent products. Thank you to @leerob from @cursor_ai, @bcherny of @claudeai code, @EnoReyes of @FactoryAI, @silasalberti of @cognition, @0xine of @semgrep, Rajesh Bhatia of @Cloudflare , @amasad of @Replit, and @eladgil.
All resources will be available online. All classes will be available to the public.
9/22 on Stanford campus. See you in class.
themodernsoftware.dev/
Sanka Mohottala retweeted
I have discovered the the BEST way to get 3D modeling results from LLMs
The app LiDAR scans the room while taking hundreds of high quality photos.
Then a workflow + skill I've created models out every item recursively checking the model against the photo at the exact spot + angle where the photo was captured
The result is a a dimensionally accurate and high quality model of your room
All in my latest video on how to do it youtube.com/watch?v=JGQCcCzu…
Sanka Mohottala retweeted
Book, Essential Mathematics for Convex Optimization by Fatma Kılınç Karzan & Arkadi Nemirovski
www2.isye.gatech.edu/~nemiro…
Introduction to Linear Optimization by Arkadi Nemirovski
www2.isye.gatech.edu/~nemiro…
Statistical Inference via Convex Optimization by Anatoli Juditsky & Arkadi Nemirovski
www2.isye.gatech.edu/~nemiro…
Sanka Mohottala retweeted
Our book "The Elements of Differentiable Programming" can now be read directly in the browser:
diffprog.github.io/