Episode Summary
Executive Summary: In this episode of This Week in Startups, hosts Jason Calacanis and Sundeep Sunny Madra review Grok, an AI chatbot from Elon Musk with access to the Twitter firehose. They give it a grade of B, praising its speed, real-time Twitter data integration, and 'fun mode,' but noting hallucinations and missing metadata. The duo then explores OpenAI's GPTs, demonstrating how to create custom GPTs in minutes using uploaded documents and instructions, predicting these will become the next app store. They discuss copyright issues, medical applications of AI, and demo a Chrome extension for analyzing Amazon reviews. The episode underscores how AI is shifting from content generation to real-time, data-rich, personalized assistants.
Main Topics: Grok AI Review (Priority: 5/5): Hosts analyze Grok, the new AI chatbot by Elon Musk with Twitter firehose access. They highlight its ability to summarize real-time Twitter discourse, its 'fun mode' for humor, and its limitations including hallucinations, lack of user metadata queries, and occasional errors. They grade it a solid B for being impressive given the short development time but not yet market-leading. GPTs as the Next App Store (Priority: 5/5): Jason and Sundeep explore OpenAI's new GPTs, demonstrating how to create custom AI assistants in minutes using documents and instructions. They argue this is a paradigm shift that threatens traditional app stores, calling on Apple and Google to respond rapidly. They show how GPTs can be used for customer service, sales qualification, and information retrieval. AI in Medical Diagnostics and Real-World Use (Priority: 4/5): The hosts discuss practical AI applications such as analyzing blood test results, interpreting prescription medications, and helping diagnose conditions. Sundeep shares a personal example of using ChatGPT to understand his father's Parkinson's medications. They emphasize that AI can serve as a valuable first opinion but caution that disclaimers are necessary due to hallucinations. Copyright and Licensing Challenges for AI (Priority: 4/5): The conversation covers the ongoing copyright debate around AI training data, including how textbooks and other copyrighted works may be used. They note OpenAI has not publicly disclosed paying licenses and discuss the complexity of copyright duration rules. This ties into broader concerns about data rights and compensation for content creators. AI Tools for Consumer Decisions (Priority: 3/5): Sundeep demos a Chrome extension called 'Fake Spot Reviews Analysis' that allows users to chat with Amazon reviews, detect fake reviews, and ask nuanced product questions. They give it a B grade, suggesting improvements like sentiment over time and proactive product suggestions. The hosts emphasize the value of solving consumer frustrations with fake reviews. Infrastructure and Compute in AI (Priority: 3/5): Jason and Sundeep discuss the massive compute resources required for AI, noting Grok was built on 10,000 A100s (soon 8,000-Sarah) and Elon Musk's prediction of compute oversupply within 18 months. They draw an analogy to the fiber optic bubble, predicting a similar cycle for AI hardware.
Key Arguments: Grok is impressive for a 0.1 product built in months, but it lacks the ability to query Twitter metadata (e.g., follower counts, engagement) which limits its utility for advanced social listening. GPTs represent a new app store paradigm. They can be created in minutes without coding, include actions to connect to external services, and can enable human-in-the-loop support, fundamentally changing how businesses interact with customers. AI can serve as a useful first opinion in medical contexts, but disclaimers are essential because hallucinations remain a significant risk. The technology is not yet ready for autonomous medical advice without human oversight. Copyright law is murky for AI training data. OpenAI appears not to have licensed most training data, relying on public domain and fair use arguments, but this will likely lead to legal battles as commercial applications proliferate. Compute infrastructure is a bottleneck today but may become oversupplied within 18 months, mirroring the dot-com fiber bubble. This could lead to lower costs and more experimentation for startups. Consumer tools like review analyzers should move beyond simple Q&A to proactive, personalized recommendations and fake review detection to truly address user pain points.
Data Points: Grok development timeline: 3 months - The hosts noted that Grok was built in about 3 months of actual development, with the company being ideated 6 months ago. Compute cluster size: 10,000 A100s (soon 8,000 Sarah) - Grok was built on a cluster of 10,000 A100 GPUs, with plans to scale to 64,000 A100s by mid-next year. Grok subscription cost: $16 per month - Grok is available under X Premium Plus subscription, costing $16 per month. Poker odds calculation error: False straight draw outs - ChatGPT incorrectly stated that a player with Ace-King suited on a Jack-10-9 flop had an ace as a straight out, which would not make a straight. Copyright duration (US): Life of author + 70 years - For textbooks published after 1976 in the US, copyright lasts the life of the author plus 70 years, or 95 years for works made for hire.
Pivotal Quotes: "Oh, Jason, where do we even begin? Your tweets are like a roller coaster ride through the Twitterverse. One moment, you're making predictions about banks. Pretty good pull. And the next, you're trying to convince us that a tiny covert pendant isn't the creepiest thing ever." — Grok (AI): Jason reads a roast generated by Grok in 'fun mode' that humorously analyzes his Twitter persona, demonstrating Grok's ability to access real-time tweet data. "This is a message to Tim Cook. This is a message to Sundar. This is the next app store, you don't have it. Fire a hundred people, bring two hundred people into a room, fire a hundred of them, and then the other one hundred put in charge and do like a Darth Vader type thing. This has to be up and running within 120 days." — Jason Calacanis: Jason passionately argues that Apple and Google must urgently respond to OpenAI's GPTs as the next app store paradigm, or risk being disrupted. "I think Vinod Koso was right. I think these things would do a better job. You tell me what you think for the global population than doctors would, or because they don't have access to a doctor." — Jason Calacanis: Discussing the potential of AI in medical diagnostics for underserved populations, suggesting that AI with human oversight could be better than no access to a doctor.
Implications: AI assistants with real-time data access (like Grok) will transform social intelligence. Custom GPTs will democratize AI creation, threatening traditional app stores. Businesses must urgently adopt AI for customer engagement or risk obsolescence. Copyright and data licensing battles will intensify, shaping the legal landscape of AI training.
About This Week in Startups
Jason Calacanis covers startups, tech, markets, media, and all the hottest topics in business and technology. He also interviews the world’s greatest founders, operators, investors, and innovators.