🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

*Task Diversity Shortens the ICL Plateau* Sometimes, learning more is easier. Training on a diverse set of in-context learning (ICL) tasks is *faster* than training for a single ICL task. (1/8) arxiv.org/abs/2410.05448

It's finally here! Excited to share the project I led with KRAFTON and NVIDIA. The future of gaming is here 🙌

🚀our paper got in #ICLR2025! Rare-to-Frequent: Unlocking Compositional Generation Power of Diffusion Models on Rare Concepts with LLM Guidance TL;DR: Exposing frequent concepts using LLMs can boost the compositionality of diffusion models—boosting #FLUX,#SD3 using #GPT! 🧵👇

🚨Seeking for test-time search/scaling of diffusion? Check our work accepted by #ICLR2025! Alignment without Over-optimization: Training-Free Solution for Diffusion Models TL;DR: Aligning rewards (Aesthetic or CLIPscore) of #SD15 or #SDXL in sampling steps without fine-tuning

🚀Rare-to-Frequent (R2F) just got #spotlight at ICLR'25! Truly believe it can pave the way to solve the "running out of data" problem as it generates Rare/OOD concept compositions without compromising image quality!

🚀Diffusion Alignment as Sampling (DAS) also got #spotlight at ICLR'25! New Title (*title changed) - Test-time Alignment of Diffusion Models without Reward Over-optimization Code & Paper links remain the same

🚀 Introducing DAS (Diffusion Alignment as Sampling) — a test-time method that aligns diffusion models using rewards without over-optimization! (Summary 🧵👇) 🔥 This work will be presented as a 𝐒𝐩𝐨𝐭𝐥𝐢𝐠𝐡𝐭 at #ICLR2025

📢Meet Smart Zoi – our new experimental feature that brings Zois to life with personality and intelligence!💡 Check the recommended specs to enjoy it to the fullest. For more details, please refer to the announcement ➡️ playinzoi.com/en/news/8419 #inZOI #SmartZoi #NVIDIA

Super excited to finally see inZOI EA released! In addition to its beautiful graphics and gameplay, I bet this game will go down in history as "the first commercial video game to truly showcase what AI is capable of" 🙂 Here are the top 3 AI features ... 🧵

inZOI NVIDIA AI And while writing this, InZOI just hit #1 on Steam globally — huge congrats to the inZOI team and KRAFTON AI for their relentless hard work to make this happen 😍

Recently, there's growing interest in LLM agents solving video games 🎮! But did you know ... about TWO years ago, we designed a GPT-4 agent that can solve the detective game HerStory (faster than humans) 😆? Details here: krafton-ai.github.io/sherlock/ Krafton AI

T1 Tool-integrated Self-verification for Test-time Compute Scaling in Small Language Models

Even a small language model can benefit from test-time scaling by self-verifying its responses with the aid of external tools! Nice work by KRAFTON AI’s Minki Kang, Jongwon Jeong, and Jaewoong Cho! Also, I love the title's homage to T1 LoL 😁

📝 Accepted at #ICML2025 Awesome job done by Junhyuck, and always thankful for Dimitris Papailiopoulos’s guidance :)

Kinda cute that you can reduce KV cache by replacing it with a universal, transferable dictionary + old school sig. proc reconstruction algorithm. We tested on non-reasoning models and was sota, but methinks it'll work even better on reasoning ones. The ICML random coins

As a video gaming company, Krafton AI has secretly been cooking something big with NVIDIA AI for a while! 🥳 We introduce Orak, the first comprehensive video gaming benchmark for LLMs! arxiv.org/abs/2506.03610

🚨New Paper Alert🚨 Excited to share our new video game benchmark, "Orak"! 🕹️ It was a thrilling experience to test whether LLM/VLM agents can solve real video games 🎮 Looking forward to continuing my research on LLM/VLM-based game agents with Krafton AI !

🚨New Paper Alert As a game company, KRAFTON AI is actively exploring how to apply LLM agents to video games. We present Orak—a foundational video gaming benchmark for LLM agents! Includes Pokémon, StarCraft II, Slay the Spire, Darkest Dungeon, Ace Attorney, and more in🧵

Orak🎮 benchmark leaderboard is just launched! Submit your LLMs and agentic strategies to compete in diverse real-world video games! krafton-ai.github.io/orak-leaderboa… *Orak comes from 오락, a native Korean word meaning “game”