🔥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

(ICML 2025 Spotlight-Top 2.6%) Multi-agent LLM systems still fail—but who caused it, and when? 🔥 We introduce Who&When, the first benchmark for automated failure attribution in LLM multi-agent systems. We also define a new task: automatically identifying which agent failed, and

Even just for organizing and orchestrating complex prompt-based systems with no optimization, DSPy is absolutely wonderful to use

Ilya had a really good slide about how “success is guaranteed” if you have: 1. Big model 2. High quality data 3. Lots of (2) And that defined the next few years of ML It is now going to be something like: 1. Big model 2. Lots of diverse inputs 3. High quality reward signal

Moving forward I think RL papers should be reporting results with random rewards as a baseline instead of just the base model

Headphones on Chipotle ordered GPUs at max usage Arbor features being added Folks, this is as good as it gets

👀👀

New short course: DSPy: Build and Optimize Agentic Apps DSPy is a powerful open-source framework for automatically tuning prompts for GenAI applications. In this course, you'll learn to use DSPy, together with MLflow. This is built in partnership with Databricks and taught by

Before LLMs, people were very precise when discussing the inputs/outputs of ML systems and the flexibility of LLMs have allowed us to become a little lazy in this department

fwiw "slop_in -> slop_out" is a valid DSPy signature for all tasks if you really want to avoid having structure

Seattle bros get real quiet when you bring up the topic of air conditioning

Maxime Rivest 🧙‍♂️🦙 Are you perhaps looking for our own Noah Ziems's Arbor? It can be used with or without DSPy, and it has the perfect OpenAI-compatible client-server architecture that works well for LLM inference. github.com/Ziems/arbor