It's finally here: The public (and most complete) version of my talk covering every stage of the process to build Olmo 3 Think. This involves changes and new considerations of every angle of the stack, from pretraining, evaluation, and of course post-training.

Here's a collection of useful patterns I've found after vibe-coding 150 different single-file HTML tools over the past couple of years simonwillison.net/2025/Dec/10/ht…

Simon Willison Nice! Here's how I serve my tools on GH pages: gist.github.com/intellectronic…

I made that because now if a Python package doesn't have good release notes I can click "copy diff" and paste into Claude/ChatGPT/Gemini and get genuinely useful release notes a few seconds later No need to even add a prompt, the models guess that's what you want from the diff

Simon Willison This is going to be really useful! I have been adding some tools here: mimansajaiswal.github.io/tools/ but I am going to adopt some of the practices that you have mentioned :D

Simon Willison Wow, I independently came up with the same idea/conclusion (that all my "scripts" should simply be 1-file HTML "apps", with CSS & JS). As an example, ashwinmenon.com/finance/stock_…. Though, I am nowhere close to the 150 you have (maybe 10 for me so far).

Stanford literally dropped a 70-minute masterclass on how GPT works

he listed the top repos and resources for learning low-level GPU programming and working with inference engines. if you love LLMs, you’ll love this too.

Check out their blogs if you are into AI/ML. 1) Andrej Karpathy Neural networks & LLMs explained from first principles by one of the OGs of modern AI. - karpathy.ai/?utm_source=li… 2) Sebastian Raschka, PhD Deep dives into LLM training and fine-tuning with super clear code

This course literally shows how to design scalable systems in under 1 hour

Open-sourcing AutoGLM, a vision-language model that understands phone screens and acts as an autonomous mobile agent. - Get started: github.com/zai-org/Open-A… - Why we built it: xiao9905.github.io/AutoGLM/blog.h… - Model weights: huggingface.co/zai-org/AutoGL… You can also access the model via API

> Find a paper on arXiv. > Build a shitty version of it. > Open source it on GitHub. >Write a post explaining why it's shitty. That GitHub repo is 100x more valuable than any certificate.

cursor is starting to look a lot more like figma 🤔

GPT-5.2 evals

just found this 512GB DDR5 RAM kit in my mailbox, anyone missing a RAM order?

> no loyalty anymore > every new release they switch models & agents > one day they swear by Gemini 3 Pro > next day they’re preaching Opus 4.5 like it’s the chosen one > then they roll up with GPT-5.2 like nothing ever happened > like crews don’t matter anymore > disgraceful

i caved & upgraded to ChatGPT Pro > GPT-5.2 Pro is one of the smartest entities i ever interacted with > gonna spend sometime today updating my local agents harness to make it use GPT 5.2 for planning and Opus 4.5 in execution > i give it ~8 months before local models catch up

this is an example of what i was talking about

calling it now, bookmark this for later - opensource AI will win - AGI will run local, not on someone else’s servers - the real ones are learning how it all works > be early > Buy a GPU > get ur hands dirty > learn how it works > you’ll thank yourself later it’s gonna be great