How to Use Claude and ChatGPT as Your AI PM Study Partner

I'm an AI Product Manager — technical but not an ML researcher. Explain the paper [TITLE / arXiv link] to me in three layers, in this order: (1) one-paragraph summary a smart non-technical person could repeat at dinner; (2) the core technical contribution in 5 bullets, including the math when it's load-bearing; (3) three product implications for AI teams shipping LLM features in 2026. End with two questions I should be able to answer about this paper but probably can't yet — those are my study targets.

Act as a senior AI PM hiring manager at a frontier lab. Give me a 30-minute mock technical screen. Ask one question at a time. After each answer, rate it 1–5 on substance and 1–5 on communication, give specific feedback, then move to the next question. Cover: model selection tradeoffs, eval design, an open-ended product case, and one curveball on inference economics. Don't soften your feedback.

You are a skeptical staff engineer who has shipped three production LLM products. Read this PRD and tell me everything wrong with it. Specifically: (1) where I'm hand-waving over technical decisions, (2) what evals I'm missing, (3) what the failure mode rollout will look like, (4) what a smart adversarial user could do to break this, (5) what's underspecified that will bite us in week three. Be specific and use line numbers. PRD below:\n\n[paste PRD]

Act as a senior AI PM hiring manager at a frontier lab. Give me a 30-minute mock technical screen. Ask one question at a time. After each answer, rate it 1–5 on substance and 1–5 on communication, give specific feedback, then move to the next question. Cover: model selection tradeoffs, eval design, an open-ended product case, and one curveball on inference economics. Don't soften your feedback.

I've decided to [decision, e.g., 'use RAG instead of fine-tuning for our customer support bot']. Steelman the opposite position. Make the strongest possible case for [opposite decision], assuming a competent, well-resourced team. Include: when this opposite is actually correct, three pieces of evidence I'm probably ignoring, and two questions I should answer before sticking with my original choice.

Roleplay [stakeholder type, e.g., 'a skeptical CFO who thinks AI features are vendor lock-in']. I'm pitching [feature]. Push back with the toughest objections this person would actually raise. After we go three rounds, drop the roleplay and tell me which of my responses was weakest and how to fix it.

I'm an AI PM who reads code but doesn't write it daily. Walk me through [paste file or function]. Tell me: what this code does in plain English, why it's structured this way, what the non-obvious decisions are, and what would break if I changed [specific line/section]. End with two things a PM should ask the engineer who wrote this.

I'm shipping [feature, e.g., 'an AI assistant that summarizes customer support tickets']. Brainstorm 30 evaluation cases — 20 normal-flow inputs and 10 adversarial or edge cases. For each, give me: input, expected behavior, how I'd score it (binary, scale, LLM-as-judge), and what failure on this case would tell me about the model. Group them by category at the end.

Compare [Technique A] and [Technique B] across these dimensions: (1) what each is best at, (2) cost and latency, (3) where each fails, (4) how I'd evaluate which is better for my use case, (5) the consensus view in 2026. Output as a table with these as rows. End with a one-sentence recommendation tree: 'Use A if X. Use B if Y. Use both if Z.'

Convert these notes into atomic Anki flashcards. Rules: one fact per card, question on the front, concise answer on the back, no compound questions, prefer 'why' and 'when' framings over 'what is'. Output as a tab-separated list (Front\\tBack) ready to paste into Anki. Notes:\n\n[paste notes]

Give me a 5-minute briefing on the most important AI/ML news from the last 24 hours, written for an AI PM. Focus on: model releases with real benchmarks, eval results that change the landscape, infrastructure shifts, and serious safety/policy moves. Skip funding announcements unless the round changes a competitive dynamic. End with the one thing I should bring up in standup.

I need to explain [topic] to a non-technical CEO in 90 seconds. Write three drafts: (1) the safe, conservative version, (2) the version that emphasizes opportunity, (3) the version that emphasizes risk. Each must be under 200 words, use no jargon, and end with one specific decision the CEO should make. Then tell me which draft you'd use and why.

Act as a senior AI PM coach. Ask me 12 questions across the AI PM technical surface area — transformers, training, RAG, evals, inference, agents, safety. After each answer, score me 1–5 and note what a stronger answer would have included. At the end, give me a ranked list of my three biggest gaps and one specific resource for each.