🚀The return of BERT, LLMs that beat physicians?!, and the dawn of agents to simulate users. Presenting Santa's nice list for AI! 🎅
- ModernBERT
- Superhuman performance of LLMs against Physicians (discussion with author @adamrodmanmd.bsky.social )
- LMagent: A Large-scale Multimodal Agents Society
- ModernBERT
- Superhuman performance of LLMs against Physicians (discussion with author @adamrodmanmd.bsky.social )
- LMagent: A Large-scale Multimodal Agents Society
Comments
-MetaMorph: Multimodal Understanding and Generation via Instruction Tuning
-AniDoc: Animation Creation Made Easier
-Adaptive Computation Modules
- Cultural Evolution of Cooperation among LLM Agents
- The Open-Source Advantage in Large Language Models
- LlamaFusion
Learn more about these papers below📷 👇
ModernBERT introduces a state-of-the-art encoder-only transformer optimized for speed, memory efficiency, and long-context tasks.
Evaluates OpenAI’s o1-preview model on complex clinical reasoning tasks, comparing its performance to human physicians and prior LLMs.
A scalable framework for simulating dynamic, multimodal multi-user behavior using large-scale multimodal LLMs.
A black-box method that jailbreaks AI models across text, vision, and audio by applying simple prompt augmentations. BoN reaches high success rates, like 89% on GPT-4o and 78% on Claude 3.5 Sonnet, with 10,000 augmented prompts, and bypasses defenses like circuit breakers.