Build With Us:Testing AI Systems After Andrew Ng|Chat vs Gemini vs Claude+Follower Paradox Challenge скачать в хорошем качестве

Build With Us:Testing AI Systems After Andrew Ng|Chat vs Gemini vs Claude+Follower Paradox Challenge

In this session, we turn Andrew Ng’s philosophy into practice: build with us, but only after you’ve practiced and thought things through. We’ll explore how to work systematically with ChatGPT-style models, Gemini, and Claude, discuss how to test the AI systems you build, and even tackle a fun logical puzzle about followers and networks. We’ll start by framing a “Practice, then Try” workflow: instead of randomly poking at models, we’ll outline a deliberate way to prototype with Chat, Gemini, and Claude so you can compare their strengths on the same task. Along the way, we’ll introduce a challenge question: In a group of people, is it possible for everyone together to have more followers than the people they follow? We’ll use this as a mini case study for reasoning, prompting, and verification. Then we’ll move into one of the most important and under-discussed topics: How do you test the AI systems that you build? We’ll look at ideas like structured test cases, golden datasets, and automatic evaluation runs—plus a discussion of tool ideas such as using an embedded OpenAI browser-style agent for testing (and the very real practical issue: do you now need a new Mac?). We’ll also brainstorm alternative testing strategies and tools you can use today, regardless of hardware. What We’ll Cover 🧪 Build With Us (After Andrew Ng): Practice, Then Try Why “watch course → immediately ship product” is backwards. A better loop: learn → practice small tasks → compare models → then build a system. How to apply this loop across Chat, Gemini, and Claude on the same problem. 🤖 Chat vs Gemini vs Claude: Practical Comparison How to design a fair “build with us” experiment across: ChatGPT-style models (“Chat”) Google’s Gemini Anthropic’s Claude Example tasks: reasoning, coding, planning, explanation. What to look for: consistency, honesty about uncertainty, and style differences. 🧩 Challenge: Follower Paradox Question “In a group of people, is it possible for everyone together to have more followers than the people they follow?” How to turn a logic puzzle into an AI evaluation scenario. Prompting models to reason step-by-step instead of hallucinating an answer. Using puzzles like this to probe reasoning, edge cases, and model reliability. 🧱 How to Test the AI Systems You Build Why testing AI is different from testing normal software—but still essential. Techniques to consider: Golden test sets & canonical answers Scenario-based test prompts Regression testing over time as models change “Red team” prompts to stress-test behavior How to combine automatic checks with manual review. 🧭 Testing Tool Ideas: OpenAI Browser & Beyond Idea: using an OpenAI browser-style agent to run automated test scenarios. Practical blocker: needing specific hardware (e.g., a new Mac) and what that implies. Other ideas and tools you can use today without upgrading machines: Scripted notebook tests API-based evaluation scripts Lightweight dashboards for tracking scores and failures over time. Resources (Add links here as you publish resources, for example:) Practice Materials & Challenges: (Add link to your practice/challenge repo or doc) AI Testing Templates / Scripts: (Add link to your testing examples, if any) Host: Mark Kerzner –   / markkerzner   ElephantScale Webinars: https://elephantscale.com/webinars/ Keywords Andrew Ng, Build With Us, AI Practice, ChatGPT, Gemini, Claude, AI Model Comparison, Logic Puzzle, Follower Paradox, AI Reasoning, Testing AI Systems, AI Evaluation, Golden Datasets, OpenAI Browser, AI Tools, Agentic Testing, Mark Kerzner, ElephantScale, Weekly AI Webinar. If you want to build AI systems that you can actually trust, hit the subscribe button and click the bell 🔔 so you don’t miss our upcoming hands-on, test-driven AI sessions.