Sasha Dichter

Subscribe

Category: AI

  • Using AI to Make Chicken Tortilla Soup from the Taco Project

    I’ve written before about how AI is a 90% expert: smarter than 90% of the people on most topics. And that was a year ago, so I’m sure the number has inched up.

    The question is not how good the AI is, the question is: what do you use it for?

    There are millions of industrial-use applications: at 60 Decibels, we’ve been cooking up artifacts in Claude like they’re going out of style; our engineering team is in overdrive; and timelines for our Impact Ecosystem Map and for our Signal product keep getting faster.

    But if you’re not already on the AI train, all of that might feel a few too many steps ahead.

    So let’s talk about cooking instead.

    For a while, my wife and I have wanted to replicate the chicken tortilla soup from the Taco Project. It’s a little bit smoky, has a deep flavor and an intriguing mix of spices. The thing is, they’ve not put the recipe online, no one seems to have posted about it, and there are hundreds of chicken tortilla soup recipes out there.

    This weekend, ChatGPT and I set out to solve this problem.

    Here were my prompts, in order:

    • I want to make chicken tortilla soup like the taco project
    • Is this better or worse in an instant pot
    • Ok let’s do it normal. I have homemade broth. Make me a shopping list
    • Can I get them [dried guajillo peppers] at Whole Foods [aside: no, I went to El Mercato instead]
    • Ok I want you to look a bit harder for the taco project version [it found customer reviews of the soup and asked me a handful of questions about those reviews, to which my responses were along the lines of ‘It’s pretty red and tomatoey. It’s not like chicken broth’]
    • Quickly compare this to other top rated recipes online and explain the difference
    • Ok let’s incorporate Rick Bayless’s technique—give me the instructions with that incorporated
    • Are you sure I’m just adding the chicken breast whole? That seems not great how will it end up w the right texture

    This set of discussions got me to a finished soup, and this is where it got interesting. The soup had potential, it looked right, and the texture was good, but it didn’t taste right at all, so we started talking again.

    • It’s close. It was tasting kind of acidic and flat. I feel like I cooked it a lot. I added salt because the homemade broth has very, very little. Probably a full teaspoon, maybe more, and I added another half a lime’s juice. It’s brightening up. I will say the Taco Project one has a creaminess to it, and I’m just trying to figure out what to do because it’s generally good. It’s close, but it’s not there yet.
    • [It asked some more questions, one of which I replied:] The Taco Project one is smooth and velvety, it doesn’t seem to have cream in it

    It instructed me to cook 2 soft corn tortillas in a skillet with the broth, add that plus some olive oil to the blender, blend that to emulsify the oil, return to the pot and cook some more.

    • That helped. It’s still a little tomato acid forward.

    It suggested I add another ½ cup of homemade broth, ¼ teaspoon of sugar, and adjust the salt.

    • I added more like a half a teaspoon of sugar, another half a cup of broth. I had already added an extra half a cup of broth before.

    It reflected on my changes and asked how it tastes now (aside: I talk to ChatGPT, I don’t type)

    • Yeah, this is closer
    • The texture is right. It’s probably a little thinner, but it’s good. The flavor is just a tiny bit on the acid side, and it’s just a little flat. Doesn’t have quite as much depth. I wonder if they put more fat in it.
    • Can you give me a printable version of the recipe w all the changes I made at the end including the salt and sugar

    My reflection is: when I asked it for a starter recipe, what I got back was a slightly more convenient version of search. That is nice and helpful, but a 90% expert to find a good chicken tortilla soup recipe is not so interesting.

    But a 90% expert around “this is how my soup (a recipe I’ve never made before) tastes, this is what I want it to taste like, help me adjust it” is not something I had access to before. It’s like having a good, experienced chef in my kitchen.

    The point is: the thinner you can slice a problem, the more likely it is that your favorite AI tool will be a game changer. General expertise has been at our fingertips for decades, but advice specific to my exact situation and preferences? That’s new.

    Here’s how it came out.

     

    And here’s the recipe if you want to give it a go. I should note that I am leaning in to healthy fats so this uses olive oil which I suspect is not traditional.

     

     

  • Two jobs at once

    All the data on multitasking notwithstanding, on any project that requires real thinking, today it feels like the only answer is to do two jobs at once.

    Project 1 one involves big data dumps, Q&A, or long voice notes that Claude or your favorite tool are then processing for a few minutes at a time.

    Project 2 can be just like project 1, or it can be easier-to-move-through work of any kind.

    After years of learning to work on one thing at a time, I’m finding myself pushing on parallel paths more than I’m used to. Otherwise I’m waiting around for Claude in ways that don’t feel productive.

  • Instinct: My AI Assistant

    A couple of weeks ago I was sitting with a fellow board member the night before a Board meeting.

    We were talking about simulations—AI-powered tools that create digital twins of people who then can be surveyed. I find this topic fascinating (and a bit terrifying), and have mostly been of the view that digital twins, by definition, will tell you the 80% that was easy to find out and leave out the 20% that really matters.

    He shared that he thought there was more potential than what I was describing, and told me about a podcast with the founder of Similie.AI that I should listen to. When I asked him to send it to me, he took out his iPhone, opened up text messages, and said: “Send Sasha from our Board the Similie.AI podcast that I listed to the other day.”

    And, from that alone, I received the following email:

    Subject: Latent Space on Similie

    Worth a listen: Latent Space with Joon Sung Park on Simile – building digital twins that simulate how people behave and make decisions, now running tens of millions of simulations for Fortune 100 clients at 85-99% accuracy vs. human focus groups.

    Thought of you and 60 Decibels.

    https://www.latent.space/p/simile

    Even with all my exposure to AI, this felt a bit like magic, so I asked him how he did it and he invited me on to the beta rollout of Instinct, which I’ve been using for the last few weeks.

    Instinct is a text-based AI (meaning I use it on iMessages on my iPhone). I suspect the tech behind it is not particularly revolutionary, but the texting interface has changed my daily operating rhythm significantly.

    My first step, scary as it was, was to give Instinct access to my Gmail, calendar, and Google Drive. Then I started texting with it.

    It became clear that it could fill a gap as a very, very useful personal assistant.

    Here are a bunch of actual texts that I’ve sent to Instinct in the last week:

    • Mark my calendar as OOO (so busy for scheduling purposes) Mon to thurs next week for my trip to Europe. Decline all my internal meetings for next week besides Friday.
    • I need to remember to keep or cancel my Amsterdam Airbnb by sunday but remind me midday Friday
    • What is [Name] phone number, he’s been slow to reply on email so if I don’t hear back by tomorrow I’ll text him. Remind me.
    • What’s the latest on our conversation with [client] about [topic]?
    • Email [Name] telling him I sent the payment asking him to confirm
    • And let’s keep track of this I need to talk to [NAME] to figure out what time works, but if you’re able to see about wed afternoon that’d be great if not, I’ll do it
    • What happened with [name] who I spoke to last week? I feel like I need to nudge her
    • Remind me Thursday to write a blog post about using Instinct (!!)

    What’s been interesting is that, despite the tech being “simple” (for today’s LLMs), the form factor radically changes how I use Instinct, specifically:

    • A texting bot is just different than opening up the Claude app (I suspect this is 100% behavioral on my part)
    • Speed of responses is high, so I use it for a different set of things
    • I cannot help but personify it a bit—I didn’t realize, when I downloaded it, that my use case was going to be “personal assistant” but that happened quickly and naturally

    My main use case is:

    1. Have a thought about something I’d like to do (work or non-work related)
    2. Send a text to Instinct
    3. It gets it 95% done and asks me to confirm it can go ahead and take the action

    The net result of this is that thoughts that occur to me quickly (and fleetingly) get taken care of with almost zero effort, meaning many fewer things get put off.

    When a bunch of low-but-not-zero-effort tasks become, essentially, zero effort—accept those 5 meeting invitations; tell me all the email I’ve gotten today that looks like spam and clear them out; check on the status of X—the effect is to (1) Lift some load from me; and (2) Get a lot of small tasks taken care of; (3) More space is freed up for more meaningful things.

    All of that, together, creates a surprising amount of leverage.

    Instinct is still in Beta but I have some invites—if you’re really interested give me a shout.

     

  • EWIMO writing

    It’s become horribly easy to spot AI writing. The false tension. The emojis and em-dashes.

    A partial solution is to teach your favorite AI tool not to write like that. I have a Claude .md file that instructs it to avoid negative parallelisms, bold-first bullets, filler transitions, and too many em-dashes. This way, if my Claude were giving advice on a chicken recipe, it would never say things like:

    Not a stew. Not a soup. A braise.

    Three hours. One pot. No shortcuts.

    This isn’t dinner — it’s an argument about how you spend a Sunday. Everything from the sear to the silence at the table is part of the same tapestry.

    The lid goes on.

    And then you wait.

    While this is all humorous, it’s everywhere, and our tolerance for it speaks volumes.

    The old tells for “I didn’t care enough to do a good job on this” were typos, unfinished sentences and numbers that didn’t add up.

    Now, everything that’s error-free but screams AI sends the message: “I phoned this in, so don’t trust me.”

    The way you avoid this is by actually doing the work: rereading each word, thinking about it, acknowledging that “I sent this” means “these words faithfully represent me and my best thinking.”

    There’s no problem using the tools to help you, but you have to practice EWIMO writing, always.

    Every word is my own.

  • On AI-generated content, trust, and gatekeepers

    Today, I received, by email, a great article by Woody Allen on antisemitism.

    Except that it wasn’t written by Woody Allen, it was fake and was written by AI.

    That got me thinking about how media and publishing work.

    For a long time, we lived in a world of gatekeepers: a few network TV channels, a few trusted newspapers, and a few trusted publishing houses.

    We’ve since had a few decades of the long tail: from cable news to the internet to social media to podcasts, everyone has become a content creator as the gatekeepers’ power has faded.

    We’re now entering an era where anyone can create content that looks like it was created by someone else (a column by “Woody Allen,” a book by “Ann Patchett”), and our ability to distinguish between what’s real and fake will keep decreasing.

    How are trust and provenance established in this new world? Will the pendulum swing back to a greater role for gatekeepers?

    I’m not saying we’ll go back to four networks, or that self-published books will disappear (they’ve gone from about 100,000 a year in 2011 to more than 2.5 million in 2025).

    Rather, I’m asking: if anyone can use AI to write a column that looks like it was written by anyone, isn’t the only way to tell who wrote that column to see where it was published?

  • Learning to Work with Claude CoWork

    Since our team hackathon, I’ve been working much more with Claude CoWork. It’s amazing, to be sure, and it also takes some getting used to.  Here are some reflections.

    My biggest adjustment going from ChatGPT/Google—tools I use to answer questions—to Claude CoWork—a tool I use to do jobs—is how not-instant the turnaround is.

    I, and we, have been trained to ask a question and get an instant answer. That is not what CoWork is like at all.

    My CoWork experience is:

    • Describe a problem
    • Clarify the problem
    • Wait a while
    • Review the solution
    • Give feedback
    • Wait a while
    • Repeat a few times
    • Deploy
    • Test
    • Give feedback
    • Wait a while
    • Clarify
    • Deploy
    • Test
    • Give feedback
    • Wait a while
    • Etc

    What I’ve found surprising is how much faster the non-bolded (my input) part is than the bolded (wait a while) part. That adds up to a lot of waiting.

    While this is not inherently problematic, the experience is not “I’m going to work on this problem for the next hour.”

    It’s much more like talking to a person who’s going to work on something, having that person go away and come back 5 minutes later. It’s fast enough that you can’t really do another thing, but slow enough that you can’t only do this thing.

    For example:

    On Friday I set out to build a web app to manage our EOS deployment.

    (Aside: EOS stands for Entrepreneurial Operating System, and it’s a well-established management system used most often with small- to medium-sized companies—somewhat akin to OKRs. You can read all about it in Gino Wickman’s book, Traction.)

    We’ve been running EOS for about a year at 60 Decibels, and for each team we have a detailed Google Sheet with that team’s quarterly priorities and workplans.

    Now that we’re deploying EOS across functions, that adds up to 5-6 EOS workstreams, each with 3-5 quarterly plans, and each plan has 5-10 weekly tasks for various team members, plus other To Do’s. It quickly has become a lot of (too many) spreadsheets and tasks to manage.

    Of course there are EOS software solutions we could buy, but, at this point, the last thing I want is another piece of SaaS software that doesn’t quite do what I want.

    So I decided to build a webapp that would (1) Give a global view of EOS across the company; and (2) Help each person on the team easily see everything they need to do across all their workstreams each week.

    I set aside a chunk of time on Friday to do this, and I did build a working app over the course of a few hours. Hooray! It’s pretty great.

    But the build experience was different from what I expected: much more like baking sourdough (a small amount of doing, a lot of waiting) than writing a blog post (“I am working for the next XX minutes”).

    I’m not used to structuring time / work this way. Instead of focused work, it feels more like “fiddle with this during the commercial break” (which is what I did watching the women’s NCAA Final Four, after I had the app up and running).

    Even right now, I’m writing this blog post while tweaking a few things on the webapp in the background, because each of my inputs is <30 seconds and each output takes a few minutes.

    It’s workable, but I’m finding it difficult to bounce back and forth between more focused tasks and Claude CoWork. So I’ve been working on the webapp while dealing with my Inbox, or catching up on Slack. That’s fine, but not a great way to spend 3-4 hours.

    I’m sure some of this is my own inefficiency, because I’m not eyeballing any of the code before deploying it. And, to be clear, this is not to take away from how amazing it is that I have useful enterprise software that I’m deploying on a Monday with ~4 total hours of attention. But the experience is much more herky-jerky than I’d expected, and I’m a bit stumped with how to work these builds into my (and my team’s) regular workflows.

    And, I must confess, it’s also surprising how amazingly easy seemingly big (UI) things are, and how difficult other small-seeming problems are (‘only show me tasks that are due this week).

    (Somewhere out there, a PM and an engineer are quietly chuckling to themselves…)

    More seriously, if you’re a leader asking your teams to lean in aggressively to these tools, you’ve got to use them yourself to develop some intuition about where they are great, where they are challenging, and what it takes to go from a “wow” prototype to something your teams can actually use.

    (Bonus: if you want to geek out on the scale of change that is happening, whether there really is a SaaSPocalypse coming, and how much org charts might be changing, check out the latest Decoder (Verge podcast) with Todd McKinnon, CEO of Okta.)

    Finally, a number of small, in-the-weeds tips if you’re new(er) to CoWork:

    • Create .md files for each topic you’re going to go deep on.
    • Recommendation is <200 lines each
    • I started with 5 background files for topics I wanted to go deep on
    • I am also creating updated .md files at the end of long sessions in the hopes that Claude will learn from those sessions (my preferences, mistakes it made, etc.)
    • For any properly large project, I’m using our company Max account. I have a Pro account also for myself, but I’m finding Pro times out more than Max, and I run out of tokens pretty quickly, even for seemingly small things.
    • For small and quick things, I’m still using Projects in ChatGPT, though I’d rather not
    • I’m only using the Desktop App
    • I was told by a friend that I should actually be using Claude Code for everything, not CoWork, but I haven’t yet.

    Here’s a screenshot of my new app.

  • The Global Majority Data Project

    I’ve been quietly working on an idea that I’m very excited about. It’s an initiative that will bring more of 60 Decibels data to life and will make this data—and the lived experience of the people it represents—more useful to more people.

    The basic concept is that AI is being trained on data from a narrow slice of humanity. The voices of billions are missing.

    This initiative, called The Global Majority Data Project, will build the infrastructure to make the lived experiences of the world’s most underserved communities both vivid and accessible, so that these experience can influence how decisions are made.

    I’m in the early stages of fundraising for this work, and I’ve been talking about it with folks I know in the AI for good space.

    After a few great conversations last week, the thought occurred to me: a nice website for this initiative would help move it forward.

    But who has time to build a website? Certainly not me (time, or know-how).

    And then I thought about the AI Hackathon we’re doing this Friday at 60 Decibels and realized it was time to eat my own cooking.

    I had the idea for the website at the end of the day on Friday.

    On Saturday, I sat down at noon with Claude CoWork.

    At 3pm the site was live.

    This is a massive step change in leverage for me (and it could be for you).

    What I had on hand at noon was a whitepaper describing the project and existing branding from 60 Decibels.

    Three hours of conversation with Claude (including a decent amount of waiting, because it was ‘thinking,’ so I cleaned up the kitchen) to make a high-quality website.

    Think of what else it can do.

    For those of you who have been using these tools, this is not a surprising outcome. For those who haven’t, I hope this serves as a concrete example of what is now possible.

    Normally I’d end the post here, but I thought some specificity would help.

    If you’d like to read on, here’s a detailed summary of what I told Claude and how the whole process unfolded.

    I hope you’ll take this as inspiration to go build something meaningful.

    And if you have ideas for folks who might want to make this idea come to life, please send them my way! (the contact form is on the website).

    My Claude Prompts to build globalmajoritydata.com

    1. My initial prompt

    OK I want your help building a website to support the Global Majority Data Project. Here’s the white paper that it is based on.

    I already have a company – 60 decibels – so I don’t need this to be a new company,

    it is an initiative of 60 decibels. I am raising money for this and want to see if I can make a decent website easily. Ultimately I expect that if it’s good, the pages will move to the 60decibels.com domain, but I just purchased globalmajoritydata.com on GoDaddy.

    So what I want to do is:

    1. I’ll share the brief
    2. I’ll share a few websites that look like what I’m interested in – obviously the branding should mostly align with 60 decibels
    3. I want you to give me some options for how to approach this, ask questions, etc…I don’t want you doing too much work building until we are aligned on how it’s going to be structured and what it will look like
    4. I’ll want you to create the structure of the site and the draft copy for me to approve
    5. Once that’s sorted, we’ll work on look and feel
    6. Then you’ll do the homepage
    7. Finally, you’ll do the rest of the site

    How does that sound?

    Any questions?

    1. Claude asked two rounds of clarifying questions—tone, structure, site size, audience, etc.—which I answered. I also uploaded my existing Global Majority Data Project whitepaper.
    2. I had to fiddle with the Chrome for Claude extension, which wasn’t behaving. Claude suggested skipping it; I insisted we fix it. We did.
    3. With Step 1 done, Claude asked for reference sites (Step 2). I shared a few. It reflected back what it inferred from them and proposed two structures: single long scroll or multi-page.
    4. I reacted to the structure, added a missing page, and outlined what it should contain.
    5. Claude drafted full-site copy, page by page. I reviewed it, gave specific comments, and it incorporated them.
    6. It asked for images. I found ~30 from internal documents and from various 60dB published assets and uploaded them.
    7. It produced a first draft of the site. Some of it was jaw-droppingly ugly. I gave pointed feedback. We went ~4 rounds on design, then I did smaller copy tweaks.
    8. The site was nearly ready. Claude generated HTML pages for browser preview with no images. It said mine were too low-res, so I bundled them into a PPT and re-uploaded.
    9. A few more rounds of feedback about the images on the site: adding variety, eliminating repetition, tightening a few style choices.
    10. (Fun moment.) I said, “Looks great, let’s publish.” Claude replied, “You’ve flagged a style issue three times that I haven’t fixed. Want me to try another approach?” I said yes.
    11. I asked how to publish. It gave instructions, which I followed. I was mostly lost at this step, but it walked me through them (with me pasting a bunch of screen shots and saying, ‘what does this mean’?). I had a handful of DNS issues that took some time to resolve.

    And that was that. The site is live. Pretty cool, huh?

     

  • Turning the Corner on AI

    The context for AI has changed.

    If you haven’t yet, I’d encourage you to read Something Big is Happening, by Matt Schumer.

    It clearly explains the AI revolution taking place right in front of us. Here’s how the essay starts:

    Think back to February 2020.

    If you were paying close attention, you might have noticed a few people talking about a virus spreading overseas. But most of us weren’t paying close attention. The stock market was doing great, your kids were in school, you were going to restaurants and shaking hands and planning trips. If someone told you they were stockpiling toilet paper you would have thought they’d been spending too much time on a weird corner of the internet. Then, over the course of about three weeks, the entire world changed. Your office closed, your kids came home, and life rearranged itself into something you wouldn’t have believed if you’d described it to yourself a month earlier.

    I think we’re in the “this seems overblown” phase of something much, much bigger than Covid.

    Once you read that, I’d encourage you to look at this webinar announcing Claude CoPilot (40 mins full speed—but you can get through it in ~15 if you really want to).

    It starts with this statement (paraphrasing): “most people are using AI to ask questions; the people who are going to stand out are the ones who get it to do things.”

    I am in the category of “good at using AI to ask questions, but I don’t use AI to do things.” That’s not where I want to be so I’m going to make a new, more concerted effort to cross the next chasm—and we’re holding an AI Hackathon at 60 Decibels next week to jump start our next phase of usage of these tools. According Boris Cherny at Anthropic (from the webinar), he has CoPilot doing his project management for him, and he used it to unsubscribe from a ton of services he wasn’t using. (I appreciated the honesty of his top two use cases).

    AI isn’t going anywhere. If you sit near anyone whose job used to be writing code, you know that they’ve mostly stopped writing code and are now having AI tools do that work for them. Their leverage is 2-10x (and increasing).

    That’s where we all could be.

    And, as they say, the best time to jump in was 5 years ago, but the next best time is today.

    (friendly reminder: for all of these tools, you want the paid version. It makes all the difference).

  • The Problem with AI Writing

    “Writing and reading is a way of underscoring that human connection is important. That you can know my mind and I can know your mind — which is a vastly consoling idea…So may times in my life, I felt a more articulate version of myself emerge after a period of writing. And when that happens, the world changes.”

    – George Saunders, author of 13 books and a recipient of the National Book Foundation’s medal for Distinguished Contribution to American Letters

    George Saunders, author of Lincon in the Bardo and many other works, is an acclaimed author and essayist who has also been called “the ultimate teacher of kindness and of craft.”

    I’ve been casting about for an explanation of what’s worrying me when I read more and more AI-generated content. Saunders’ explanation—you can listen the whole conversation here—gets to the heart of the matter.

    The question is: what is writing, really? And Saunders’ answer is anchored in the idea of two people knowing each others’ minds, and the beauty in that connection.

    That’s the thing that’s tripping me up with all the AI-generated writing, a doubly confusing experience I’m also finding my own AI helper(s) to be radically useful: as brainstorming partners, as “I think this thing isn’t quite right, but I’m not sure how to make it better,” as a project planners… but very rarely as a writer, and never as a writer of this blog.

    Like George Saunders, I’m writing for two reasons: to figure something out (a sharpening of my understanding that comes from sitting with and articulating an idea), and to personally connect with readers.

    And something breaks when I’m not the person writing the words, and when I’m reading words not written by a person.

    I’m sure you’re having this experience as often as I am: working your way through an article or an email or some other document, and you come across a sentence or a phrase that is so stylistically ChatGPT that you stop…and in that moment, feel a strong sense of disconnection.

    For example, I was reading this interesting post about searching job listings and pay rates to understand what skills AI firms are paying for to feed their models (answer: “the bottleneck for AI is no longer information (facts); it is expert reasoning (process).) While the research and the conclusions in the article are interesting, the writing is so clearly authored by ChatGPT that I quickly disengaged.

    The reason for that disengagement Saunders’ severing of human connection.

    It’s as if there’s an implicit equation that I’m carrying around in my head, something about balancing:

    The author’s effort in getting the words just right =  My effort as a reader to sit with and deeply understand what the author is saying.

    Because when the author cares enough to toil over every word, that means that they are carrying around an idea that matters deeply to them.

    If the author truly cared, if the author takes the work personally, then I do too.

    (aside: if I’m reading something equivalent to a textbook. If I just am acquiring knowledge or skills, I’m fine with the computer writing most of that text).

    Whereas, if my effort in deciphering a 1,000 word post is greater than what it took the author to create it…something feels fundamentally off. I feel like I’m being had, that that essay is part of the mountain of words generated by a computer model that doesn’t mean much, really, to the person who “wrote” it.

    What’s curious is that the models will get better and this moment will pass. What happens when it’s impossible to tell the difference?

    I don’t know how to answer that with confidence, but I’ll stand by my view that I’m reading because I want to see through a window into the mind of another person—whether it’s fiction or nonfiction, analysis or some fantastical world.

    The intent and the human behind the words are the reason writing comes alive. It isn’t just words on a page, it is the effort, struggle and care of one person opening up their mind and heart. That’s why I’ve rewritten this post four times in 12 hours 🙂

    We write what we believe, what we care about, what we’ve thought about, and then we share that with others. The connection is personal, the words are the medium.

    The moment that connection is severed, I find it much harder to show up as a reader with my full attention.

  • Podcast Alert: Patrick Thean

    Patrick is an author, CEO coach and mentor, and I was happy to get the chance to join his podcast recently. Our discussion covered topics that will be familiar to readers of this blog and, as always, I discovered that talking about a topic helps me understand the topic better.

    If I had to summarize my reflections from the conversation, they are:

    1. How we do anything is how we do everything
    2. One way to learn how we do anything is by digging into new domains we are exploring

    What follows is a conversation that starts with 60 Decibels,  my motivation for doing work in the social sector and my Generosity Experiment; but quickly turns to learning to play the guitar, my favorite AI use cases, and the attitude Patrick and I would suggest towards getting over your lingering AI resistance.

    You can listen to the full podcast here (20 mins):

    • Apple: https://bit.ly/3GB5W0P
    • iHeart Radio: https://bit.ly/40eTeLU
    • Spotify: https://bit.ly/4lqkhfM