Starting off the last sessions at Interrupt we have @HeggieConnor and @kunal_rye from @unifygtm sharing tips on building research agents! Tl;dr the human component is still key – Start with a simple vibe check, then moved to human labeling
– Several axes they tweaked:
–
PROMPT ENGINEERING
-
Building Research Agents: Human Component Remains Essential
By
–
-
LangGraph Studio v2: Local Web App with Prompt Editing and Debugging
By
–
LangGraph Studio v2 We're launching a new and improved version of LangGraph Studio. Includes:
– No more desktop app! Run it locally
– Edit prompts and other configuration directly in studio
– Playground integration
– Pull down production traces and run them locally to debug -
Andrew Ng on Evals: Start Simple, Augment Human Judgment
By
–
Incredible fireside chat at Interrupt right now with the legendary @AndrewYNg ! – "I think of evals as something I throw together in 5 minutes", not a big task to start. Augment human judgement, improve over time
– People telling others to avoid learning coding will go -
Crafting AI Prompts for Research Success
By
–
One Prompt: Your Perfect Research Assistant
Crafting AI prompts improves research accuracy and clarity.
• Specify goals
• Use clear language
• Seek actionable insights Click below to read more: https://
godofprompt.ai/blog/one-promp
t-your-perfect-research-assistant
… -
OpenEvals: Simulating and Evaluating Multi-Turn LLM Conversations
By
–
How to simulate and evaluate multi-turn conversations Most LLM applications today are chat-based. How would you evaluate the conversations? We’re excited to launch OpenEvals — a set of utilities to simulate full conversations and evaluate your LLM application’s
-
LangChain Launches Private Preview Tools for LLM Judge Evaluators
By
–
We just launched in private preview new tools for better building and aligning LLM as a judge evaluators: Align LLM as Judge: based on @eugeneyan
's fantastic Align Evals work, this makes it easier to bootstrap LLM as a judge evaluators
Audit LLM as Judge: make sure your LLM -
Claude 4o Prompt Idea: Unhinged MSN Messenger Conversation
By
–
Gave me an idea for a 4o prompt: > a screenshot of an unhinged MSN messenger conversation from 2002
-
o3 Model Performance Varies With Different Instructions
By
–
Yeh I dunno if it’s some memory or instruction that veers it off. I felt like past instructions helped the models but with o3 it performs differently when changing instructions
-
Reasoning tokens alone cannot solve fundamental AI limitations
By
–
Adding more reasoning tokens is not going to fix this