By playing, you agree to uListen's Terms and Privacy Policy and YouTube's Terms of Service.
My First MillionBrainstorming ChatGPT Business Ideas With A Billionaire | ft. Dharmesh Shah (#438)
Summaries, chapters, transcripts, and FAQs on this page are made by uListen, not YouTube.
CHAPTERS
- 0:00 – 1:43
Why Generative AI Is a Paradigm Shift
Dharmesh says AI is the biggest tech shift since the internet and that it far outweighs mobile. Shaan sets up the conversation, and Dharmesh proposes starting with generative AI and ChatGPT.
- •Dharmesh calls AI the largest opportunity in his 30 years in software
- •Mobile had a discrete set of use cases, while AI affects everything
- •Shaan describes feeling like he is knitting while a party goes on next door
- •Dharmesh has been obsessed with ChatGPT since it launched
- 1:43 – 6:32
OpenAI, Sam Altman's Equity and the Nonprofit-to-Profit Story
The hosts discuss reports that Sam Altman holds no equity in OpenAI and whether it could reach a trillion-dollar valuation. Shaan recounts the reported Elon Musk funding story behind the for-profit subsidiary, and Dharmesh gives his view of Altman.
- •Dharmesh puts OpenAI in the top three candidates to reach $1 trillion
- •Shaan relays the reported story of Musk pulling funding and OpenAI creating a capped-profit subsidiary
- •Dharmesh says large language models are capital intensive for a software company
- •Dharmesh sees Altman as a reasonable, non-evil capitalist
- 6:32 – 9:55
AI Week: Kanye Rap Intro and a Website in 10 Minutes
Shaan describes clearing his calendar for an AI week of experimenting. He built a podcast rap with ChatGPT and a Kanye voice model, and then made a website with ChatGPT's help, including troubleshooting Netlify.
- •ChatGPT wrote an intro rap for My First Million
- •A Kanye voice model run through Google Colab turned Shaan's rapping into Kanye
- •He made a Hello World site and ChatGPT walked him through hosting on Netlify
- •ChatGPT diagnosed that Netlify wanted a folder, not a file
- •Shaan had two mind-blowing moments in the first 15 minutes
- 9:55 – 16:48
Text-to-Code and the Chat UX Revolution
Dharmesh explains the third generative use case, text to code, and why it enables chat-based interfaces to software. He contrasts imperative and declarative models and uses HubSpot reporting as an example. Shaan then walks through his own experience building a site with ChatGPT.
- •Text-to-code lets natural language prompts generate working code
- •Chat UX replaces clicks and drags with describing the outcome you want
- •Imperative models give step-by-step instructions, while declarative models describe the result
- •The new requirement is knowing what you want and being able to say it
- •Conversational memory lets users iterate and even have the AI interview them
- 16:48 – 19:15
Is ChatGPT Just Fancy Autocomplete?
Shaan asks whether ChatGPT is just autocomplete or something more. Dharmesh calls GPT a reasoning engine and says it goes beyond simple probabilistic next-word prediction.
- •Shaan describes the autocomplete mental model and why it doesn't match his experience
- •Dharmesh says GPT-3 and 4 are reasoning engines, not knowledge bases
- •He says reducing it to autocomplete is like calling computers just zeros and ones
- •He does not claim the model has consciousness
- 19:15 – 24:33
Is AI Scary? Optimism vs. Doomsday Bunkers
Sam asks whether Dharmesh is afraid of AI, and Shaan presses on the warnings from Musk and Altman. Dharmesh stays optimistic, framing AI as human to the AI power rather than human versus AI, and doubts billionaire bunkers are caused by GPT.
- •Dharmesh frames AI as an amplifier of human ability, like computers
- •Shaan raises the danger of misuse and uncontrolled optimization, citing the red team testing
- •Shaan mentions Altman's reported bunker
- •Dharmesh guesses billionaires were building bunkers before GPT-3
- •He says an AI takeover is not likely anytime soon
- 24:33 – 28:21
Sequoia AI Event Takeaways and ChatGPT Plugins
Dharmesh describes the Sequoia AI event, including predictions on text-to-video. He explains Sam Altman's plugin announcement, which he compares to the App Store and sees as a bigger deal than GPT-4.
- •Attendees focused on practical matters like tech stacks and what comes next
- •Dharmesh expects text-to-video, even feature films, to improve quickly
- •Plugins let third parties connect ChatGPT to the internet and proprietary data
- •He compares plugins to the iPhone App Store, making ChatGPT an ecosystem
- 28:21 – 30:29
Is This the New Internet Opportunity? Pandora and First Movers
Sam uses the Pandora story to ask whether this is another moment where even mediocre success could be huge. Dharmesh says yes, and calls AI an order of magnitude bigger than mobile and comparable to the original web.
- •Sam relays Tim Westergren's Apple access and first-mover advantage with Pandora
- •Sam asks if this is the moment to catch a tidal wave
- •Dharmesh says it's too big to ignore and bigger than mobile
- •He says it opens opportunity for startups and incumbents alike
- 30:29 – 41:26
Vector Embeddings Explained and Business Ideas
Dharmesh walks through vector embeddings, from points on a line to 1,000-dimension space, as semantic distance rather than keyword matching. He applies it to Sam's Hampton community and discusses dating, fashion and Pinecone.
- •Text can be reduced to a vector, and distance between vectors measures relatedness in meaning
- •The opportunity is to replace crude keyword matching in an industry
- •Embeddings could match Hampton members on issues like founder struggles
- •Faceted search and keyword search differ from embeddings, and Pinecone is a leading vector database
- •Dharmesh says someone could build an embedding model of a dataset in a weekend
- 41:26 – 46:23
Tools of the Trade and Plugins in Practice
Dharmesh names Python, Pinecone and LangChain as emerging tools. Shaan then gives a travel-planning example of plugins, and Dharmesh describes solving for the experience rather than the transaction.
- •Python is the common language of the AI world
- •LangChain is an open source project for chaining prompts and models
- •Shaan imagines ChatGPT booking trips or updating Salesforce through plugins
- •Dharmesh envisions an assistant that knows you and solves for the whole experience
- 46:23 – 50:25
Why Dharmesh Bought Chat.com and Built ChatSpot
Dharmesh explains that he personally bought Chat.com because he believes chat is the future interface. He describes building ChatSpot.ai largely himself, and the hosts note its simple page and 19-minute video.
- •He bought Chat.com personally for eight figures and has no plan yet
- •He funds side projects himself and says no HubSpot P&Ls are harmed
- •ChatSpot was built mostly by Dharmesh with freelancers using OpenAI's APIs
- •ChatSpot is being transferred to a HubSpot team
- •Sam notes ChatSpot's video has about 200,000 views
- 50:25 – 54:03
What Shaan and Sam Admire in Dharmesh
Shaan lists what he admires about Dharmesh: enthusiasm, tinkering, simple content and guts. Sam adds the rare combination of nerdy artistry and commercial success.
- •Enthusiasm after 30 years is the fountain of youth
- •He keeps tinkering despite running a big public company
- •His simple screen-share content shows he doesn't overthink it
- •Shaan sees the domain purchase as an ante that forces him to do something
- 54:03 – 1:00:28
Dharmesh's 17-Year Chatbot Journey and Courage of Convictions
Dharmesh traces the idea from IngeniMail to GrowthBot to ChatSpot and Chat.com, calling Chat.com his ante to get into the AI party. He cites Bill Gates on natural language interfaces and explains why he stays at HubSpot.
- •IngeniMail was an email-based assistant idea from 17 years ago
- •GrowthBot failed because natural language understanding wasn't good enough
- •GPT made the old idea possible, leading to ChatSpot
- •Chat.com is the cover charge for the AI party
- •Dharmesh advises having the courage of your convictions and iterating
- •He crafted a HubSpot role with no direct reports or one-on-ones
- 1:00:28 – 1:06:28
Advice for Sam and Hampton: Intersect or Double Down
Dharmesh advises Sam to combine his community-building skill with AI technologies like vectors. Shaan argues instead that Sam should know his nature and double down on his proven formula. Dharmesh also suggests a tweak to Hampton's application question about role.
- •Dharmesh suggests Sam intersect community building with AI and vector embeddings
- •Shaan recalls Sam resisting video platforms on The Hustle and being right
- •Shaan says to know your nature and not chase every new thing
- •Dharmesh recommends loosening the role question to avoid pushing people out early
- 1:06:28 – 1:17:15
Shaan's AI Experiments, Prompt Engineering and Closing Thoughts
Shaan explains his immersion approach to AI and his podcast-to-idea database workflow. Dharmesh then covers prompt engineering, his prompt.com purchase, a view on Balaji's Bitcoin bet, and a warning against being an AI tourist.
- •Shaan uses immersion weeks inspired by Bill Gates' reading week
- •He used Whisper and ChatGPT to extract every idea from the podcast
- •Dharmesh calls prompt engineering a new skill and has bought prompt.com
- •He doubts Balaji's odds on the dollar and Bitcoin
- •He warns against being a grifter or AI tourist
