Whether to collect preferences ("do you prefer response A or B?") from the same person who wrote the prompt, or a different person, is important and understudied. Highlighted this question in a recent talk docs.google.com/presentation/d…. Sycophancy probably results when you have the

Over the past year, I've been compiling some "alchemist's notes" on deep learning. Right now it covers basic optimization, architectures, and generative models. Focus is on learnability -- each page has nice graphics and an end-to-end implementation. notes.kvfrans.com

One fundamental issue with RL – whether it’s for robots or LLMs – is how hard it is to get rewards. For LLM reasoning, we need ground-truth labels to verify answers. We found that maximizing confidence alone allows LLMs to improve their reasoning with RL!

Excited to present FastTD3: a simple, fast, and capable off-policy RL algorithm for humanoid control -- with an open-source code to run your own humanoid RL experiments in no time! Thread below 🧵

Everyday human data is robotics’ answer to internet-scale tokens. But how can robots learn to feel—just from videos?📹 Introducing FeelTheForce (FTF): force-sensitive manipulation policies learned from natural human interactions🖐️🤖 👉 feel-the-force-ftf.github.io 1/n

🤖🌎 We are organizing a workshop on Robotics World Modeling at Conference on Robot Learning 2025! We have an excellent group of speakers and panelists, and are inviting you to submit your papers with a July 13 deadline. Website: robot-world-modeling.github.io

Excited to share Adaptive Low-Pass Guidance (ALG): a simple training-free, drop-in fix that brings dynamic motion back to Image-to-Video models! Demo videos, paper, & code below! choi403.github.io/ALG (🧵 1/7)

If you are interested in I2V generation, please check out June Suk Choi’s recent work! Simple and effective method based on deep analysis.

This is what I needed!

📢 Excited to announce the 1st workshop on Making Sense of Data in Robotics Conference on Robot Learning! #CORL2025 What makes robot learning data “good”? We focus on: 🧩 Data Composition 🧹 Data Curation 💡 Data Interpretability 📅 Papers due: 08/22/2025 🌐 tinyurl.com/corldata25 🧵(1/3)

Join our CoRL 2025 workshop on data-centric robot learning! We’re accepting submissions now 🗓 Deadline: Aug 22 🔗 tinyurl.com/corldata25

🧵When training reasoning models, what's the best approach? SFT, Online RL, or perhaps Offline RL? At KRAFTON AI and SK telecom, we've explored this critical question, uncovering interesting insights! Let’s dive deeper, starting with the basics first. 1) SFT SFT (aka hard

Modern AI is confined to the digital world. At Skild AI, we are building towards AGI for the real world, unconstrained by robot type or task — a single, omni-bodied brain. Today, we are sharing our journey, starting with early milestones, with more to come in the weeks ahead.

Self-Questioning Language Models: LLMs that learn to generate their own questions and answers via asymmetric self-play RL. There is no external training data – the only input is a single prompt specifying the topic.

We're hosting the 1st workshop on Making Sense of Data in Robotics at Conference on Robot Learning this year! We'll investigate what makes robot learning data "good" by discussing: 🧩 Data Composition 🧹 Data Curation 💡 Data Interpretability Paper submissions are due 8/22/2025! 🧵(1/3)

Just wrote a long-overdue blog post on Weave-Head Attention: a minimal change that substantially boosts training stability at scale.

The term “AGI” is currently a vague, moving goalpost. To ground the discussion, we propose a comprehensive, testable definition of AGI. Using it, we can quantify progress: GPT-4 (2023) was 27% of the way to AGI. GPT-5 (2025) is 58%. Here’s how we define and measure it: 🧵

Introducing DEAS, a scalable offline RL framework utilizing action sequences with stable value learning. 💪🏼 SOTA performance in complex tasks in OGBench. 😳 DEAS can be used to improve VLA in both simulation and real-world tasks. 🤗 Code and datasets are all open-sourced!

Simulation drives robotics progress, but how do we close the reality gap? Introducing GaussGym: an open-source framework for learning locomotion from pixels with ultra-fast parallelized photorealistic rendering across >4,000 iPhone, GrandTour, ARKit, and Veo scenes! Thread 🧵

Instead of automating what humans do, we explore AI agents that help people stay focused and follow through with intention. Introducing INA, an AI agent for intentional living (led by Juheon Choi ) I believe INA is a step toward human-centered AI that supports more mindful