Huge thanks to all the open source projects that've made a lot of the tech we rely on in the world possible:
Linux
Git
FFmpeg
PyTorch & TensorFlow
Apache & Nginx
MySQL, PostgreSQL, SQLite
Chromium & Firefox
GCC & LLVM
Docker & Kubernetes
Also, all the open-weight LLMs... and…
LLMs can make sense of retrieved context because of how transformers work.
In one of the lessons from the Retrieval Augmented Generation (RAG) course, we unpack how LLMs process augmented prompts using token embeddings, positional vectors, and multi-head attention. Understanding…
We have a very poor understanding of why deep neural networks like transformer models learn the parameters they learn. For example, in the paper below from 2013, the authors demonstrated that 5% of the weights of a trained deep neural network can be used to predict the values of…
You sometimes see this feature in ChatGPT and Gemini when it shows you two answers to the same question in parallel and asks you to choose one.
It's especially revealing when you ask for a fix for a bug and you see on the left an entirely wrong assessment of the situation and on…
Every problem whose description contains enough information to solve it and can fit into one or several pages of text can be solved by an LLM finetuned on enough examples of such problems.
Really, three years into this, and I still see even smart people falling into this "OpenAI…
Becoming an RL diehard in the past year and thinking about RL for most of my waking hours inadvertently taught me an important lesson about how to live my own life.
One of the big concepts in RL is that you always want to be “on-policy”: instead of mimicking other people’s…
I think Jack Dorsey’s new “bitchat” app makes total sense in this day and age
We’ve seen similar concepts before – Hong Kong protesters used Bridgefy during the protests for off-grid communication to avoid police surveillance
As authoritarianism rises worldwide, data centers…
I think Jack Dorsey’s new “bitchat” app makes total sense in this day and age
We’ve seen similar concepts before – Hong Kong protesters used Bridgefy during the protests for off-grid communication to avoid police surveillance
As authoritarianism rises worldwide, data centers…
Wrapped up Stanford CS336 (Language Models from Scratch), taught with an amazing team @tatsu_hashimoto@marcelroed@neilbband@rckpudi. Researchers are becoming detached from the technical details of how LMs work. In CS336, we try to fix that by having students build everything:
If you use "AI agents" (LLMs that call tools) you need to be aware of the Lethal Trifecta
Any time you combine access to private data with exposure to untrusted content and the ability to externally communicate an attacker can trick the system into stealing your data!
What if LLMs could learn your habits and preferences well enough (across any context!) to anticipate your needs?
In a new paper, we present the General User Model (GUM): a model of you built from just your everyday computer use.
🧵
Every frontier AI system should be grounded in a core commitment: to protect human joy and endeavour. Today, we launch @LawZero_, a nonprofit dedicated to advancing safe-by-design AI. lawzero.org
Agentic Document Extraction just got much faster! From previous 135sec median processing time down to 8sec. Extracts not just text but diagrams, charts, and form fields from PDFs to give LLM-ready output. Please see the video for details and some application ideas.
I’ve been fascinated lately by the question: what kinds of capabilities might base LLMs lose when they are aligned? i.e. where can alignment make models WORSE? I’ve been looking into this with @ChrisGPotts and here's one piece of the answer: randomness and creativity
Señor @NicolasMaduro, usted ha dicho en numerosas ocasiones que quiere a los venezolanos de regreso y en libertad.
A diferencia de usted, que tiene presos políticos, nosotros no tenemos presos políticos. Todos los venezolanos que tenemos bajo custodia fueron detenidos en el…
🚨New RAG Dataset Release🚨
Lead by @beirmug: we’ve curated real long and complex questions, each requiring multiple retrieved documents covering a diverse set of concepts (i.e. nuggets).
🚨New RAG Dataset Release🚨
Lead by @beirmug: we’ve curated real long and complex questions, each requiring multiple retrieved documents covering a diverse set of concepts (i.e. nuggets).
10 Followers 180 FollowingPhD student @asuengineering. Prev ECE @msfea_aub. Curious about machine learning, information theory, statistics, and some other stuff.
436 Followers 202 FollowingProfessor, Teaching Stream, UofT Comp Sci. Interested in factors for success in intro programming & effectiveness of online and inverted classrooms
10 Followers 180 FollowingPhD student @asuengineering. Prev ECE @msfea_aub. Curious about machine learning, information theory, statistics, and some other stuff.
365K Followers 8 FollowingVercel provides the developer tools and cloud infrastructure to build, scale, and secure a faster, more personalized web. Creators of @nextjs, @v0, and @aisdk.
97K Followers 899 FollowingCEO @ Civaam. Institute of Technology and Justice @UniOfOxford. Prev. AIML @Apple. Alumni @Columbia. Represented by: Emma Leong, @JanklowUK. Views my own.
571 Followers 637 FollowingDriving Growth and Innovation | Tech, Strategy | FMVA® . DataScience at MIT. Building AI solutions at https://t.co/5lh5qsBI0d .
47K Followers 110 FollowingMy new LM book: https://t.co/YXNQUy7O3t
PhD in AI, author of 📖 The Hundred-Page Language Models Book and 📖 The Hundred-Page Machine Learning Book
38K Followers 134 FollowingPhilosopher and Cognitive Scientist | Associate Professor and Award-Winning Lecturer | Get 3 chapters of our new Book for free ↓
2K Followers 2K FollowingAssistant Professor @UofT & @uoftmie. Formerly postdoc @DS4DM in Montreal and grad student @gtcomputing & @GTCSE. Machine Learning+Discrete Optimization.
2K Followers 235 FollowingAssociate Professor of Computer Science at University of Toronto. Research in human-compatible AI and large-scale studies of online platforms.
6K Followers 53 FollowingFollow us for updates/global perspectives as we invest to help provide a foundation on which Canadians can build financial security in retirement.
FR: @oirpc
6K Followers 1K FollowingThe world’s #1 telematics provider, committed to advancing technology, empowering businesses and making the roads safer for everyone! 🚛
80K Followers 25 FollowingThe platform that connects you to the money you deserve with less work and achieve financial confidence with TurboTax, Credit Karma, QuickBooks, and Mailchimp.
395K Followers 14 FollowingJoin us as we celebrate 50 years of empowering 50 million investor-owners.*
Community guidelines: https://t.co/F9iBO3PueF
*As of January 2025
1.4M Followers 1K FollowingBuilding @EurekaLabsAI. Previously Director of AI @ Tesla, founding team @ OpenAI, CS231n/PhD @ Stanford. I like to train large deep neural nets.
9K Followers 16 Following@Stanford Prof. National Acad of Eng. Chief Sci @ Visual Layer & Virtue AI. Frm Sr Dir AI @Apple. Co-author of XGBoost, LIME, TextGrad, Alpaca, TVM, GraphLab.