We just discovered the 🔥 COOLEST 🔥 trick in Flow that we have to share: Instead of wordsmithing the perfect prompt, you can just... draw it. Take the image of your scene, doodle what you'd like on it (through any editing app), and then briefly describe what needs to happen

Slides for my lecture “LLM Reasoning” at Stanford CS 25: dennyzhou.github.io/LLM-Reasoning-… Key points: 1. Reasoning in LLMs simply means generating a sequence of intermediate tokens before producing the final answer. Whether this resembles human reasoning is irrelevant. The crucial

Excited to share our #ACL2025NLP paper, "𝐂𝐢𝐭𝐞𝐄𝐯𝐚𝐥: 𝐏𝐫𝐢𝐧𝐜𝐢𝐩𝐥𝐞-𝐃𝐫𝐢𝐯𝐞𝐧 𝐂𝐢𝐭𝐚𝐭𝐢𝐨𝐧 𝐄𝐯𝐚𝐥𝐮𝐚𝐭𝐢𝐨𝐧 𝐟𝐨𝐫 𝐒𝐨𝐮𝐫𝐜𝐞 𝐀𝐭𝐭𝐫𝐢𝐛𝐮𝐭𝐢𝐨𝐧"! 📜 If you’re working on RAG, Deep Research and Trustworthy AI, this is for you. Why? Citation quality is

Maybe don't use an LLM for _everything_? Last summer, I got to fiddle again with content diversity Adobe Research Adobe and we showed that agentic pipelines that mix LLM-prompt steps with principled techniques can yield better, more personalized summaries

We're thrilled to release & open-source Hunyuan3D World Model 1.0! This model enables you to generate immersive, explorable, and interactive 3D worlds from just a sentence or an image. It's the industry's first open-source 3D world generation model, compatible with CG pipelines

Presenting LAQuer tomorrow (Monday) from 11:00 to 12:30 at poster session 1!

🚨 New paper alert! 🚨 We propose an IQ Test for LLMs — a new way to evaluate models that goes beyond benchmarks and uncovers their core skills. Think: 🧠🤖 psychometrics for LLMs. 👇 (1/6)

BIU NLP Itai Mondshine Reut Tsarfaty Our paper "Beyond N-Grams: Rethinking Evaluation Metrics and Strategies for Multilingual Abstractive Summarization": arxiv.org/pdf/2507.08342

Are you still around Vienna? Come hear about a new morphological task at CoNLL at ~11:20 (hall M.1) Reut Tsarfaty

Can LLM agents really understand us? 🤔 Paper: arxiv.org/html/2507.2203… Code/Data: github.com/SalesforceAIRe… We introduce UserBench: a gym environment testing how well agents align with nuanced human intent—not just follow commands. Users are messy: vague, evolving, indirect.

Hearing from a lot of folks that they still fine-tune Qwen2.5 instead of Qwen3 — simply because “it’s easier to tune.” Qwen2.5 models seem more steerable: easier to adapt for new behaviors or boost specific capabilities, which means more downstream work builds on them. People

Producing reasoning texts boosts the capabilities of AI models, but do we humans correctly understand these texts? Our latest research suggests that we do not. This highlights a new angle on the "Are they transparent?" debate: they might be, but we misinterpret them. 🧵

Wow! Chinese lab Tencent Hunyuan has released an open source alternative to Genie 3 🔥 You can generate realistic videos that you can control in real time. - Long-term consistency - No need for expensive rendering - Trained on 1M+ gameplay recordings Already available ↓

🚨 New preprint out! CRISP: Persistent Concept Unlearning via SAEs LLMs often encode knowledge we want to remove. CRISP enables persistent, interpretable, precise unlearning while keeping models useful & coherent—tested on bio & cyber safety tasks🧵👇 📄arxiv.org/abs/2508.13650

🚨 New paper out! 📄 What happens when LLMs & RLMs face conflicting answers to a question? 🤔 They often ignore disagreement and confidently pick one “correct” answer. 🤯 📄 arxiv.org/pdf/2508.12355 #AI #LLM #NLP #MachineLearning

How it works? [4/n] 🔹 UQ-Dataset: 500 challenging, diverse unsolved questions sourced from Stack Exchange. They are super hard and arise naturally when humans seek answers. 🔹 UQ-Validator: No ground-truth → traditional metrics fail. Instead, we designed compound validation