Out of Our Depth

Out of Our Depth

By Mary and JoshTechnology
Download on the App Store

Out of Our Depth episodes

  • The Jensen Huang Scandal, Explained Simply

    Jensen Huang ran NVIDIA for 33 years. On November 12 he resigned. We sat down to sort out what happened, what NVIDIA does, and whether we'd want to work there now.


    We split the story three ways: what he's admitted, what's only been reported, and what nobody knows yet. Then we get into GPUs in plain English, the interim CEO, the money going in circles between chipmakers and AI labs, and OpenAI's chip named Jalapeño.


    Would you take a job at NVIDIA right now, or at a competitor? Tell us in the comments.


    Chapters

    00:00 A very grounded week

    01:08 What Jensen actually admitted

    02:49 Affairs or China: why he's out

    06:26 What NVIDIA actually does

    09:10 The board, the staff and no successor

    10:55 Colette Kress and the money loop

    16:02 AI that improves itself?

    17:36 Can Colette Kress pull it off?

    21:02 Would we work at NVIDIA?

    25:44 A big L, and why leaders matter

    27:56 Back at Denny's


    Out of Our Depth

    Website: https://outofourdepth.com/

    YouTube: https://www.youtube.com/@OutOfOurDepthPod

    Instagram: https://www.instagram.com/OutOfOurDepthPod/

    Spotify: https://outofourdepth.com/spotify

    Questions or ideas: [email protected]


    What's confirmed and what isn't, as of November 14, 2026

    Admitted: on November 8, Jensen Huang told NVIDIA staff he had been unfaithful to his wife. He resigned on November 12 and CFO Colette Kress became interim CEO. NVIDIA gave no reason.

    His message: https://nvidianews.nvidia.com/news/a-message-to-our-company

    The resignation: https://techcrunch.com/2026/11/12/nvidias-jensen-huang-resigns-chipmaker-names-cfo-colette-kress-as-interim-ceo/

    Timeline: https://techcrunch.com/2026/11/13/nvidia-ceo-jensen-huangs-affair-with-alleged-chinese-spy-a-timeline/


    "Deer in the headlights" comes from a Wall Street Journal report on an August meeting.


    NVIDIA was the world's most valuable company until October 30, when Apple passed it.


    On Anthropic: its November 2 filing says future models may "imminently" reach the company's own threshold for AI speeding up AI research, and that no current model has been designated as crossing it.


    These are our opinions and our read of public reporting, recorded Saturday, November 14, 2026. The story is still moving, and nothing here is investment advice.

    32 min
  • How Bad Is AI for the Environment?

    Does asking ChatGPT a question use a whole bottle of water? That sent us down a rabbit hole about what it takes to run AI, and who pays when a data center moves in next door.


    We talk through the difference between a quick chat and an agent working for hours, why a cheaper prompt can still mean more total demand, and what happens when a new data center needs power and water. Then Mary puts Josh in the hot seat. How would he sell her on having one next door?


    Try What Uses More? Compare two digital tasks, then change the power source and see how the estimate moves.


    00:00 Does using AI make me a horrible person?

    00:41 Three ways to think about impact

    01:52 What a single prompt uses

    02:30 Why images, video and agents differ

    05:26 Training versus everyday use

    06:45 Who pays for local power?

    11:22 What happens in a real community?

    15:27 Power sources and efficiency

    16:07 What heavy use looks like

    17:16 What choices matter in practice?

    25:09 Where AI might help

    27:24 A calculator you can try


    A few numbers in context

    - Google's few-drops estimate is for its median Gemini text prompt and on-site cooling water. The older study behind the bottle comparison also models water used to generate electricity. The task, location, and scope change the answer.

    - Google's energy comparison covers its median Gemini Apps text prompt from May 2024 to May 2025. Its overall electricity use grew about 37% from calendar year 2024 to calendar year 2025. The periods overlap, and the total covers the company.

    - The IEA estimates data centers worldwide used about 1.5% of global electricity in 2025. That covers all data centers.

    - The Ireland and Virginia examples are rough comparisons across different years. Ireland's CSO reported 23% of metered electricity in 2025. EPRI estimated about 26% of Virginia's electricity supply in 2023. Both cover data centers as a whole.

    - The Dalles example refers to municipal water use. The city's 2024 records put Google's usage at about 434 million gallons, roughly a third of that year's city total.

    - Josh's usage figure is a rough estimate from his own Claude activity. Treat it as an example of a heavy workflow.


    Links and sources


    University of Maine, What Uses More?

    https://what-uses-more.com


    International Energy Agency, Key Questions on Energy and AI

    https://www.iea.org/reports/key-questions-on-energy-and-ai


    Google, Measuring the Environmental Impact of Delivering AI at Google Scale

    https://arxiv.org/abs/2508.15734


    Google, 2026 Environmental Report

    https://sustainability.google/reports/google-2026-environmental-report


    Li et al., Making AI Less Thirsty

    https://arxiv.org/abs/2304.03271


    Ireland Central Statistics Office, Data Centres Metered Electricity Consumption 2025

    https://www.cso.ie/en/releasesandpublications/ep/p-dcmec/datacentresmeteredelectricityconsumption2025


    Pew Research Center, U.S. data centers' energy use, citing EPRI

    https://www.pewresearch.org/short-reads/2025/10/24/what-we-know-about-energy-use-at-us-data-centers-amid-the-ai-boom


    Oregon Public Broadcasting, The Dalles city water records

    https://www.opb.org/article/2026/01/23/the-dalles-mayor-data-center-google

    31 min
  • How to Use AI as a Student Without Getting Dumber

    How to use AI as a student without outsourcing the thinking you came to school to learn.


    We share the routines we use at Stanford, like Mary's audio-first class prep and Josh's short primers. Claude grilled Mary more than our classmates did. Then we talk about whether AI shortcuts hurt learning, what school is for, and whether we'd choose a school built around two hours of AI a day.


    Try this: before you open AI for your next assignment, ask whether you care about doing the thinking or just the output.


    What's one thing you won't hand off to AI? Tell us in the comments.


    Chapters

    00:00 AI for busy students

    01:31 Does AI help us learn?

    03:46 The shortcut tradeoff

    06:51 Mary's audio-first routine

    09:54 Rehearse and get grilled

    11:24 Josh's short class primers

    13:00 Notes and student organizations

    14:48 Practice the cold call

    16:51 The weekly AI audit

    17:46 Keep the thinking you want to learn

    20:45 What is education for?

    22:29 Why fundamentals still help

    26:24 Oral exams and an AI professor

    29:16 Would we choose an AI school?

    34:54 Cake mix or from scratch


    Out of Our Depth: https://outofourdepth.com

    Instagram: https://www.instagram.com/outofourdepthpod/

    Run Your City: https://www.runyourcity.org

    AI@GSB: https://www.gsb.stanford.edu/experience/learning/applied-ai


    Go deeper

    The math study (Bastani et al., PNAS, 2025): https://doi.org/10.1073/pnas.2422633122

    Claude Skills Tutorial: Teach AI Your Voice and Style: https://www.youtube.com/watch?v=V_DVa4fi3q0

    Claude/ChatGPT Skills, Projects & Connectors, Explained Simply: https://www.youtube.com/watch?v=d-ykZl4WHjo


    Tools we mention

    Claude: https://claude.ai/

    Gemini Notebook (formerly NotebookLM): https://notebook.google.com/

    Speechify: https://speechify.com/

    Granola: https://www.granola.ai/

    Notability: https://notability.com/

    Alpha School: https://alpha.school/the-program/


    Questions or episode ideas? [email protected]

    38 min
  • Claude Skills Tutorial: Teach AI Your Voice and Style

    Our step-by-step Claude skills tutorial: ask Claude for three versions, tell it what feels like you, and save that feedback in a skill you can reuse.


    Mary rehearses her AI workshop on Josh, her practice audience. She shows how to teach Claude your writing voice and design taste with real examples and specific feedback. Then Josh tries it himself and stays up until 2 a.m. turning his personal site into a mini video game. Asking for options helped. The video game may have been overkill.


    New to this? Start with How to Make AI Actually Sound Like You, linked below. Try the tutorial and send us what you build. We'd love to see it.


    Chapters

    00:00 Mary's workshop dress rehearsal

    03:46 Why AI output feels generic

    08:15 Teach Claude your voice with a skill

    13:36 Teach Claude your design taste

    19:29 Plan first, then give feedback that sticks

    23:43 When to split your skills

    27:28 Try the tutorial

    29:44 Josh tries it (and overbuilds)


    Try it yourself

    Mary's step-by-step tutorial: https://aigsb-workshop3.vercel.app/

    Workshop slides: https://aigsb-workshop3.vercel.app/deck

    Mary's personal website: https://mary-blankemeier.vercel.app/

    Josh's personal website: https://josh-eiland.vercel.app/

    Searching for Birds, by Nadieh Bremer: https://searchingforbirds.visualcinnamon.com/


    Watch Mary's full workshop (bonus episode): https://www.youtube.com/watch?v=yhVYY1N7hDg


    Tools and resources

    Claude: https://claude.ai/

    Create a custom Claude skill: https://support.claude.com/en/articles/12512198-how-to-create-custom-skills

    Humanizer skill: https://github.com/blader/humanizer

    Winning Writing skills: https://github.com/kalyvask/winning-writing

    Run Your City: https://www.runyourcity.org


    Related episodes

    How to Make AI Actually Sound Like You: https://www.youtube.com/watch?v=_d-hzc6Lt5c

    Claude/ChatGPT Skills, Projects & Connectors, Explained Simply: https://www.youtube.com/watch?v=d-ykZl4WHjo


    Out of Our Depth: https://outofourdepth.com

    Instagram: https://www.instagram.com

    Questions or episode ideas: [email protected]

    35 min
  • [BONUS EPISODE] The Art of Personalization: Teaching AI YOUR voice & style

    Why does everyone's AI sound the same, and how do you make yours sound like you? In this live AI@GSB workshop at Stanford GSB, Mary Blankemeier shows how to teach AI your voice and your visual style with skills: give it examples, the vibe you want, and your icks, then keep updating the skill every time you correct it.Build your own with the tutorial: https://aigsb-workshop3.vercel.appQuestions, ideas, feedback? Drop them in the comments or email [email protected].0:00 Who I am0:53 The plan: five questions1:28 Which website would you trust?2:50 Why AI drifts to the average4:27 When it feels AI-made, trust drops6:09 What's a skill?7:30 Your voice: examples, vibe, icks8:21 Skills get good by iterating9:36 Update the skill every time you correct it10:43 Q&A: keeping only the parts of your voice you like12:35 Your design: same formula15:15 The payoff: one brand skill, a whole platform18:02 Rapid fire: what I wish I knew sooner20:31 Split skills that don't travel together21:35 Context, explained with an intern and a messy desk22:31 My favorite skills23:51 Q&A: what if your writing already sounds like AI?24:48 Go build

    26 min
  • Will AI Take Our Jobs? What AGI Actually Means

    Will our jobs still exist when we graduate? We get into what artificial general intelligence (AGI) actually means, how close we think we are, and what it could change about work and everyday life.


    Josh thinks we could have AI that feels like AGI by the end of 2027. Mary wants it to prove it can handle a real person's breadth of tasks, depth of judgment and reliability. Naturally, the conversation turns to whether it can even be trusted with her inbox.


    We unpack the "jagged frontier," why the goalposts keep moving, whether AI can improve itself, and what benchmarks can tell us. Then we get into the harder questions: what does adapting look like when you're already working and raising a family? Who gets a say in how this technology develops? And what would a future worth being optimistic about actually look like?


    Plus, Mary's vision for a world with more opportunity, a debate about work and purpose, and a one-year challenge involving AI, friendship and FaceTime.


    Try this: pick a task you know well. Decide what reliable, end-to-end help would look like, then test an AI tool against that standard.


    What would convince you we've reached AGI? Tell us your test in the comments.


    Chapters

    00:00 AGI and our jobs

    01:06 What would count as AGI?

    04:01 The moving goalposts

    05:15 Can AI improve itself?

    07:27 Our AGI predictions

    09:10 Why email is harder than it looks

    13:48 Can tests settle the debate?

    15:17 Reasons to stay optimistic

    18:13 Who needs a voice in AI?

    19:44 The downside case

    21:51 Adapting when you have no spare time

    25:50 Staying curious in uncertainty

    26:59 Who should get access?

    28:43 The future we want

    33:40 Our one-year test


    Out of Our Depth: https://outofourdepth.com

    Instagram: https://www.instagram.com/outofourdepthpod/

    Run Your City: https://www.runyourcity.org

    AI@GSB: https://www.gsb.stanford.edu/experience/learning/applied-ai


    Go deeper

    Google DeepMind - Levels of AGI: https://deepmind.google/research/publications/66938/

    ARC Prize / ARC-AGI benchmarks: https://arcprize.org/


    Questions or episode ideas? [email protected]

    37 min
  • How to Learn AI Without Feeling Behind

    Feeling behind on AI? So are we.


    We're building an AI "couch to 5K": a manageable way to understand the language, get comfortable with tools like ChatGPT and Claude, and form our own views on the bigger questions.


    In this episode, we compare what's helping us learn, from explanations built around things we love to small weekly experiments. Mary brings the girly pop energy. Josh confronts his newsletter overload and an app that probably didn't need to be an app.


    Try this with us: pick one AI term you keep hearing, ask ChatGPT or Claude to explain it through something you love, then use it on one small, real task this week.


    At the end, we show you our new Learning tab, where you can browse AI terms, vote for topics you want us to cover, and suggest anything we've missed.


    CHAPTERS

    00:00 Ask the question anyway

    02:42 From zero to one

    05:32 Fluency, agency and perspective

    08:16 Build an information diet you will use

    10:47 Make the learning stick

    12:42 Build your fluency base

    14:46 Agency: start with one useful thing

    17:19 Solve the problem before building the app

    20:53 Perspective: choose questions that matter

    27:55 What is known, unknown and worth exploring

    30:32 Fact-check and listen to the stakes

    31:57 Keep us honest: measure progress

    32:47 Explore the Learning tab and vote


    EXPLORE WITH US

    Website: https://outofourdepth.com

    Learning tab and topic requests: https://outofourdepth.com/learning


    RESOURCES MENTIONED

    Ben's Bites: https://www.bensbites.com

    Ethan Mollick / One Useful Thing: https://www.oneusefulthing.org

    tldraw: https://www.tldraw.com

    Gallup's April 2026 survey on Gen Z and AI: https://news.gallup.com/poll/708224/gen-adoption-steady-skepticism-climbs.aspx

    35 min
  • AI Planned Our Wedding

    We planned our wedding with AI. Mary gave Josh a trial period... if it got stressful, we would hire a real wedding planner.

    In this episode, we walk through what we actually used it for, from comparing venue packages and building a detailed task tracker to figuring out which linens, hair and makeup felt right. Mary brings the design judgment. Josh gets very into the logistics. Both of us get a little excited about planning another wedding.

    We also share how we built a wedding weekend website with schedules, photo sharing and a few jokes for our guests. Then we talk about using voice memos to organize memories before writing our own vows.

    Try this: give AI the details of your next big project, ask what information is missing, and have it turn the plan into tasks with owners and deadlines. Bring examples of what you like when a decision is hard to picture.

    What's one part of wedding planning you'd want help with? Tell us in the comments or email [email protected].

    Chapters
    00:00 Planning our wedding with AI
    02:30 Find the right venue
    03:58 Compare venue packages and costs
    06:09 Questions for the venue tour
    08:01 Build a wedding task tracker
    11:23 Give AI your wedding preferences
    13:35 Keep your planning context up to date
    16:45 Visualize the venue and decor
    20:47 Find hair and makeup that feel like you
    23:53 Build a wedding weekend website
    25:50 Make the weekend easier for guests
    30:11 Connect the website to your guest list
    31:01 Organize memories for your vows
    32:04 Keep wedding planning fun

    Out of Our Depth: https://outofourdepth.com
    Run Your City: https://www.runyourcity.org
    AI@GSB: https://www.gsb.stanford.edu/experience/learning/applied-ai

    Tools we mention
    ChatGPT: https://chatgpt.com
    Claude: https://claude.ai
    Gemini: https://gemini.google.com
    Granola: https://www.granola.ai
    Canva: https://www.canva.com
    Pinterest: https://www.pinterest.com
    Zola: https://www.zola.com

    Related episode
    Projects, Connectors & Skills... Take Your AI from Fresh Intern to Full-Time Hire: https://www.youtube.com/watch?v=d-ykZl4WHjo

    #AI #Podcast #OutOfOurDepth

    34 min
  • How to Make AI Actually Sound Like You

    In this episode, we work through how to make AI’s writing and designs reflect your voice, taste, and brand.


    Mary shares how she built Run Your City’s design system by asking for options, reacting to examples, and refining what felt right. Josh walks through teaching AI his writing style with real emails, Slack messages, and feedback, then adapting those instructions for different audiences.


    We also get into AI’s role in Mary’s painting process, borrowing visual inspiration, and how we use reusable skills in our grant applications while keeping people involved in the final review.


    Try this: ask your AI for three versions, explain what you like and what feels off, and turn those choices into instructions you can reuse.


    Chapters

    00:00 Personal site, personal style

    01:05 Make AI reflect your brand

    05:41 Design that feels like you

    07:33 From design system to reusable skill

    09:44 How to get started

    11:48 AI in Mary's creative process

    15:46 Borrow inspiration

    17:31 Write for the person reading

    19:14 Teach AI your writing style

    25:05 Audience, format and feedback

    28:20 Our grant-writing workflow

    32:11 Keep the human review


    Out of Our Depth: https://outofourdepth.com

    Run Your City: https://www.runyourcity.org

    AI@GSB: https://www.gsb.stanford.edu/experience/learning/applied-ai

    Ideas for a future episode: [email protected]


    Tools and resources

    Winning Writing skills: https://github.com/kalyvask/winning-writing

    Humanizer skill: https://github.com/blader/humanizer

    Claude Design: https://claude.com/product/design


    Try it yourself

    Set up a design system in Claude Design: https://support.claude.com/en/articles/14604397-set-up-your-design-system-in-claude-design

    Create your own Claude skill: https://support.claude.com/en/articles/12512198-how-to-create-custom-skills


    Related episode

    Projects, Connectors & Skills… Take Your AI from Fresh Intern to Full-Time Hire: https://www.youtube.com/watch?v=d-ykZl4WHjo

    34 min
  • ChatGPT as Our Running Coach: What We Had to Fix

    We need to prepare for El Cruce, a three-day mountain race. Mary wants a coach, cute gear and the logistics handled. Surely AI can sort all that out… right?


    We turn a messy brainstorm into ChatGPT project instructions, compare voice with dictation, and build a training website we actually want to use. The first version needs work. So do our instructions.


    Along the way, we figure out how to give AI useful context, bring examples of what we like, ask for a plan before it starts building, and stop when something is good enough. Plus a few reminders that learning AI together can get chaotic.


    Try this on your own project: give AI your goals and constraints, ask what information is missing, then request three possible output formats before it starts building.


    00:00 Our adventure and the AI brief

    04:35 Turn a brainstorm into useful context

    09:09 Save instructions and test voice

    12:23 Ask for a plan before it builds

    14:14 What the AI asks before planning

    16:23 Your taste belongs in the prompt

    18:19 Bring examples the AI can use

    20:06 Review the first result

    21:56 Define what good enough means

    23:52 A website we want to use

    25:18 Share the website

    27:15 Now for the logistics

    28:32 Better thinking beats more prompting

    31:52 Good enough and learning together

    34 min

About Out of Our Depth

From the publisher's feed

Out of Our Depth is us learning AI out loud... breaking down new developments, testing the tools, and asking the silly questions so you don't have to.