the best way to get good at something is usually to just practice actually doing the thing in question. a lot of very capable people outsmart themselves with complex plans that involve working a lot on fake prerequisites.

Good read #AI

Small matter of Virginia Tech pride! Google Scholar turns 20 today 🎉🎉🎉 Kudos to its creators, Anurag Acharya and Alex Verstak. Fun fact: Alex was my second MS thesis graduate at Virginia Tech Computer Science! ✨ See these nice articles covering their work ⬇️ 1. blog.google/outreach-initi… 2.

Vamos, Rafa Nadal!   As you get ready to graduate from tennis, I’ve got a few things to share before I maybe get emotional.   Let’s start with the obvious: you beat me—a lot. More than I managed to beat you. You challenged me in ways no one else could. On clay, it felt like I

If you're asking a question to which no one so far has the right answer, a prerequisite to correctly answering it is not accepting any of the readily available wrong answers. The belief that "others have already done it" is what stops many capable people from coming up with

Today we're announcing the winners of ARC Prize 2024. We're also publishing an extensive technical report on what we learned from the competition (link in the next tweet). The state-of-the-art went from 33% to 55.5%, the largest single-year increase we've seen since 2020. The

Google Scholar has an AI-powered PDF reader that gives you outlines of research papers. It's totally free, but many folks still don't know about it. Here's how to use Google Scholar's PDF reader: (If you bookmark this post, please like it too.)

It was a great experience going through the LLM Agents course. The course covered quite a good breadth of topics from LLM agent frameworks, software, and security. llmagents-learning.org/f24 Here's a link to a summary article from one of the lectures - link.medium.com/JQJ5xKMYhPb

I am teaching a class on #FoundationalModels for #robotics and Scaling #DeepRL algorithms. This class expands on last year's class and my generalist robotics policies tutorial and code. I plan to share the lectures and code assignments. Starting with the first lectures below.

Ken's results were shocking. Our latent space safety controller works like magic to prevent you (teleop-ing) from spilling an open bag of Skittles during manipulation! It even lets you shake the bag side-to-side, but slows you down so nothing falls out! Check the vids out in 🧵:

Summary in case you missed any LLM research from the past month: * RL on math datasets improves math ability v1 * RL on math datasets improves math ability v2 * RL on math datasets improves math ability v3 * RL on math datasets improves math ability v4 * RL on math datasets...

We're seeing the same form-versus-function confusion with Deep Research now that we saw in the early days of chatbots. Back then people were wowed by chatbots' conversational abilities and mimicry of linguistic fluency, and hence underappreciated the limitations. Now people are

Amazing !!🥳🥳🥳

Current LLM scratchpads look like human reasoning, but do they help because of human-understandable content, or are LLMs encoding thoughts in the syntax? We measure performance after distilling paraphrased scratchpads and show that encoded reasoning is NOT playing a large role 🧵

Test of Time Winner Adam: A Method for Stochastic Optimization Diederik P. Kingma, Jimmy Ba Adam revolutionized neural network training, enabling significantly faster convergence and more stable training across a wide variety of architectures and tasks.

Amusing outcome of "vibe coding" wave with AI (by people who have not written software professionally before) Seeing the speedrun of why software development REALLY is hard (it was never about writing code - even though most non-dev folks assumed it must have been!)

The health indicator no one talks about = the strong desire to work and build things. Whether for the intrinsic love of the work, the rewards, or both. Yes we need sleep and (some need) recreation but drive is at least as important as any other metric.

Nice series of seminars on LLM Agents topics ranging from reasoning to theorem proving as part of the Advanced LLM Agents course (llmagents-learning.org/sp25). Here's a brief look at Multimodal AI agent topics - tinyurl.com/5t3c2m98

Because AI is an engineering discipline and not a scientific field, it's never possible to fully separate the properties of a given approach from those of its specific implementations. The artifact is the method.

#planning #RL #AI