🔥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

Feedback Loops With Language Models Drive In-Context Reward Hacking Shows that feedback loops can cause in-context reward hacking, where the LLM at test-time optimizes an objective but creates negative side effects in the process arxiv.org/abs/2402.06627

Can hazardous knowledge be unlearned from LLMs without harming other capabilities? We’re releasing the Weapons of Mass Destruction Proxy (WMDP), a dataset about weaponization, and we create a way to unlearn this knowledge. 📝arxiv.org/abs/2403.03218 🔗wmdp.ai

My first PhD paper!🎉We learn *diffusion* models for code generation that learn to directly *edit* syntax trees of programs. The result is a system that can incrementally write code, see the execution output, and debug it. 🧵1/n

Model developers try to train “safe” models that refuse to help with malicious tasks like hacking ...but in new work with Jacob Steinhardt and Anca Dragan, we show that such models still enable misuse: adversaries can combine multiple safe models to bypass safeguards 1/n

🚀 Just Launched: VideoArena!🎥 Discover head-to-head comparisons of video clips generated from the same prompts across top text-to-video models. Compare outputs from 7 leading models and we're adding more soon! 🔗 Check out the leaderboard: videoarena.tv #Text2Video

In a new preprint, we show that VLMs can perform cross-modal tasks... ...since text ICL 📚, instructions 📋, and image ICL 🖼️ are compressed into similar task representations. See “Task Vectors are Cross-Modal”, work w/ trevordarrell, Amir Bar. task-vectors-are-cross-modal.github.io

✨New preprint: Dual-Process Image Generation! We distill *feedback from a VLM* into *feed-forward image generation*, at inference time. The result is flexible control: parameterize tasks as multimodal inputs, visually inspect the images with the VLM, and update the generator.🧵