Arize AI

??? Tickets are LIVE for Arize:Observe! ??? The premier event for AI engineers, researchers, and industry leaders is back. Join us in San Francisco on June 25 at SHACK15 for a full day of insights, discussions, and networking—all focused on AI evaluation, observability, and the next generation of agents and assistants. ?? Learn from experts tackling AI's biggest challenges ?? Explore cutting-edge techniques for evaluating AI agents & assistants ?? Connect with industry leaders shaping the future of AI As AI systems become more autonomous and high-stakes, staying ahead with rigorous evaluation methods is essential. Don’t miss this deep dive into the future of AI observability. ?? Get your tickets: arize.com/observe-2025

The?ReAct framework?combines?Reasoning + Action?to allow LLMs to not only?“think”?but also refine selected actions. In this prompt optimization tutorial, we leverage ReAct principles to prompt LLMs to?Reason + Act?just like humans do. With these fundamentals, we can instruct the LLM to dynamically reason, adjust actions, and generate “thoughts”—all within the prompt itself—giving us deeper insight into the choices the LLM is making. Watch the full tutorial to see this in action: https://lnkd.in/e6GZKfsp

Want to learn how to seamlessly transition an AI agent from text to voice? On 4.10, Sally-Ann DeLucia will talk to Brookes Stephens & Byro Mitchell about Priceline's journey in evolving Penny, their AI-powered travel assistant. They'll dive into the technical hurdles of integrating audio into complex AI systems, covering... ?? Best practices for evaluating and optimizing multimodal performance. ?? Challenges in real-time audio processing and natural language understanding. ?? Performance measurement, including key metrics and evolving evaluation strategies. Get a behind the scenes look at the innovative design and evaluation processes driving the next wave of agents. Join us live: https://lnkd.in/gx7ZKiRR

The Agent event of the week!

Arize:Observe is back! Ensure reliable performance and confident scalability for your AI agents with a full day of expert talks + insights. ?????? Here's everything you need to know about this year's event... ?? Happening on June 25 at SHACK15. Register: https://lnkd.in/gPxyfCtn ?? Talks cover the latest in agent + assistant evaluation and observability! Last year we had builders, researchers, and engineering leaders from Anthropic NATO Microsoft NVIDIA Mistral AI Lowe's Companies, Inc. Google + more. ?? The first round of speakers will be announced soon ?? ?? Right now tickets are $100, but that will increase soon! Further discounts available when you join the Arize slack community: https://lnkd.in/gVCYY_vH ?? Lunch + happy hour provided ?? Working on an app you want to demo? We have a space for that: https://lnkd.in/gYFaF2p2 ?? Want to present? We're still accepting speaker applications for another few weeks. Get in there: https://lnkd.in/gJu-sFKp

Take control of your agents. ?? We're hosting a virtual workshop on 4.15 to help you streamline agent performance through optimized prompt engineering (few-shot, meta, gradients, Bayesian). We'll cover everything from conceptual foundations to practical, UI-based and technical workflows. Come for the knowledge, stay for the...knowledge! Register here: https://lu.ma/prompt-opt

?New integration?? ?Significantly improve LLM reliability and performance with Phoenix + @CleanlabAI's Trustworthy Language Model (TLM). TLM automatically identifies mislabeled, low-quality, or ambiguous training data—ensuring models are built on trustworthy foundations Phoenix provides deep observability to debug, evaluate, and enhance LLM performance in production How it works: 1?? Extract LLM traces from Phoenix and structure input-output pairs for evaluation. 2?? Use Cleanlab TLM to assign a trustworthiness score and explanation to each response. 3?? Log evaluations back to Phoenix for traceability, clustering, and deeper insights into model performance. ?? Dive into the full implementation in our docs & notebook: Documentation: https://lnkd.in/e7aSFNuC Notebook: https://lnkd.in/egUUk4RP?

The way we prompt LLMs has a significant impact on their reasoning and problem-solving abilities. One effective approach is?Chain of Thought (CoT) prompting, which guides models to think step-by-step and break down complex problems logically. Here are three key CoT techniques to consider: 1. Standard CoT?for structured reasoning 2. Self-Consistency CoT?for more reliable outcomes 3. Few-Shot CoT?to improve performance with minimal training examples Check out the full video here to see these techniques in action: https://lnkd.in/eK7THK9n

In case you missed it, Arize AI Phoenix crossed the 5k GitHub star mark last week! ?? Phoenix has changed a TON since its first iteration. Just in the past 9 months, the team has added Prompt Management, Prompt Playground, Sessions, Experiments & Datasets, Annotations, Cloud Instances, Authentication and User Access, and dozens of auto-instrumentor updates. I'm constantly in awe of the execution speed and quality of this team. Here's to the next 5k and beyond! Shoutout to Mikyo King, Xander Song, Roger Yang, Dustin Ngo, Anthony Powell, Francisco Castillo

LLMs can solve complex problems, but how much of their reasoning is influenced by the way we prompt them? In the next segment of our prompting series, we explore Chain of Thought (CoT) prompting—a powerful technique that promotes step-by-step thinking, guiding LLMs to break down problems into logical steps. By applying various CoT methods—Standard CoT, Self-Consistency CoT, and Few-Shot CoT—we can significantly enhance an LLM’s problem-solving abilities. Watch the full tutorial here: https://lnkd.in/etiAD88C