Sebastian Li🇨🇦

I love Statistics. Ex @Amazon, ex healthcare researcher @CAMHnews, @SickkidsNEWS, PhD student at University of Toronto @UofT. Toronto-Seattle -Saskatoon-Beijing

[Mihail Eric @mihail_eric]

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. 

https://t.co/wTokHyUMsz

[Diyi Yang @Diyi_Yang]

Follow along here if you are interested 😊 We will keep this site updated with assignments and lectures!

https://t.co/68kDTefxHc

[Kirk Borne @KirkDBorne]

#Probability and #Statistics — The Science of Uncertainty (Second Edition): https://t.co/V9EL0UQg7x
(🌟774-page downloadable PDF + 369-page Solutions Manual PDF🌟)
————
#DataScientist #DataScience #ML #MachineLearning #Analytics

[Sebastian Li🇨🇦 @SebastianLi]

lol. Mom, you forgot to give me the candy!

[Sara Hooker @sarahookr]

lol the modern version of "you forgot to cite me"

[Sebastian Li🇨🇦 @SebastianLi]

@EdgarDobriban I wonder what would happen if you tried Gemini 3 Pro and other thinking LLMs.

[BURKOV @burkov]

One of the most important papers of 2025, and probably of the past three years.

DeepSeek R1 has removed the moat from OpenAI and given a reasoning LLM secret to humanity.

R1 made the reasoning LLM the de facto standard of the modern language model. DeepSeek has shown how one can take a pretrained language model that doesn't have any practical problem-solving skills, and make it learn, on its own via reinforcement learning, to solve the hardest scientific and coding problems.

Thanks to DeepSeek, we now have Claude 4.5, Gemini 3, and Kimi K2. This paper, published in Nature, explains how R1 was built.

Let the article talk to you on ChapterPal: https://t.co/g6A2Xpgcu3

Download the PDF: https://t.co/QHG350jV9T

[Sebastian Li🇨🇦 @SebastianLi]

For MATLAB you referred to, I recalled that might be Geoff’s NN Graduate Course about 20 years ago that I sat in. PyTorch in comparison made impossible at the time possible and easy today.

[Andrej Karpathy @karpathy]

Great run Soumith! Of all the deep learning framework transitions I've been through where I re-wrote ~all of my code (matlab -> caffe -> numpy -> torch -> pytorch), the PyTorch one was most pleasant and now significantly longest lasting. It hit a jackpot of the time in the 20-dimensional design space of objectives and constraints. May you find another golden era in a space that most excites you!

[Andrej Karpathy @karpathy]

Excited to release new repo: nanochat!
(it's among the most unhinged I've written).

Unlike my earlier similar repo nanoGPT which only covered pretraining, nanochat is a minimal, from scratch, full-stack training/inference pipeline of a simple ChatGPT clone in a single, dependency-minimal codebase. You boot up a cloud GPU box, run a single script and in as little as 4 hours later you can talk to your own LLM in a ChatGPT-like web UI.

It weighs ~8,000 lines of imo quite clean code to:

- Train the tokenizer using a new Rust implementation
- Pretrain a Transformer LLM on FineWeb, evaluate CORE score across a number of metrics
- Midtrain on user-assistant conversations from SmolTalk, multiple choice questions, tool use.
- SFT, evaluate the chat model on world knowledge multiple choice (ARC-E/C, MMLU), math (GSM8K), code (HumanEval)
- RL the model optionally on GSM8K with "GRPO"
- Efficient inference the model in an Engine with KV cache, simple prefill/decode, tool use (Python interpreter in a lightweight sandbox), talk to it over CLI or ChatGPT-like WebUI.
- Write a single markdown report card, summarizing and gamifying the whole thing.

Even for as low as ~$100 in cost (~4 hours on an 8XH100 node), you can train a little ChatGPT clone that you can kind of talk to, and which can write stories/poems, answer simple questions. About ~12 hours surpasses GPT-2 CORE metric. As you further scale up towards ~$1000 (~41.6 hours of training), it quickly becomes a lot more coherent and can solve simple math/code problems and take multiple choice tests. E.g. a depth 30 model trained for 24 hours (this is about equal to FLOPs of GPT-3 Small 125M and 1/1000th of GPT-3) gets into 40s on MMLU and 70s on ARC-Easy, 20s on GSM8K, etc.

My goal is to get the full "strong baseline" stack into one cohesive, minimal, readable, hackable, maximally forkable repo. nanochat will be the capstone project of LLM101n (which is still being developed). I think it also has potential to grow into a research harness, or a benchmark, similar to nanoGPT before it. It is by no means finished, tuned or optimized (actually I think there's likely quite a bit of low-hanging fruit), but I think it's at a place where the overall skeleton is ok enough that it can go up on GitHub where all the parts of it can be improved.

Link to repo and a detailed walkthrough of the nanochat speedrun is in the reply.

[Sebastian Li🇨🇦 @SebastianLi]

GPU Clusters from lightning AI. @lightningai https://t.co/aXCO3Ih51z

[Sebastian Li🇨🇦 @SebastianLi]

@JohnPasalis I am not sure 30- young Canadians would save this tax exempted. rather spend it.

[Geoffrey Hinton @geoffreyhinton]

I think Elon Musk should be expelled from the British Royal Society. Not because he peddles conspiracy theories and makes Nazi salutes, but because of the huge damage he is doing to scientific institutions in the US. Now let's see if he really believes in free speech.

[Sebastian Li🇨🇦 @SebastianLi]

Does it make perfectly political sense for the Liberals for QC cities vs your suggested ON cities? Yeah, I was also thinking about Kingston over Peterbourough.

[Sebastian Li🇨🇦 @SebastianLi]

@MikePMoffatt @SeanFraserMP @MikePMoffatt I have a high respect for your research, but the way you presented how you measured Fraser’s impact is puzzling - that BC and Ontario the two largest housing markets have actually declined should warrant an ‘FAIL’ on Fraser’s record.

[Sebastian Li🇨🇦 @SebastianLi]

Isn’t it good or bad news?

[Daniel Foch @danielfoch]

Canadian Permanent Resident Applications Fall 57%, Weakest Month In Years https://t.co/mm6G89QVpm

[Sebastian Li🇨🇦 @SebastianLi]

Income tax reduction is visible and quick to implement whereas affordable housing is a generational problem.

[Daniel Foch @danielfoch]

An overwhelming majority of Canadians feel reducing income taxes is the highest economic priority, even above affordable housing:

[Sebastian Li🇨🇦 @SebastianLi]

Do you have any stats to show us? Just for the PhD engineers (relaxing the thermodynamic part). Or just anecdotally?

[Lazy Canadian Entrepreneur @JimChuong]

There are literally PhD engineers out there who find statistical thermodynamics easy but carry a credit card balance and hold most of their TFSA in cash.

[Sebastian Li🇨🇦 @SebastianLi]

Yeah, a double standard (very normal for politicians).

[Sebastian Li🇨🇦 @SebastianLi]

Who is he? NIPS no longer NIPS 10-20 years ago

[Tongzhou Wang @ssnl_tz]

Director & chair of CIFAR, supposedly leader of our field.

Yelling at organic changes he didn’t like, blatantly attacking an entire group of AI researchers, as if he dictates the field.

Many proper ways to propose change. Screaming “I don’t like it” is childish and is not one.

[Zhijing Jin @ZhijingJin]

If you want to do a PhD with me on Multi-Agent LLMs (see our recent #NeurIPS2024 paper on GovSim https://t.co/rLTKFbMzDB), I highly recommend you apply by *Nov 14* to Cooperative AI PhD Fellowship! Up to $40K/year w/ other benefits. Email me if you want to iterate your proposal.

[Sebastian Li🇨🇦 @SebastianLi]

@danielfoch You have the source of that 20k difference to show? I couldn’t find it from any public free sources (hence I suspected).