Episode Summary
Executive Summary: In this episode of This Week in Startups, Jason Calacanis and Sandeep Nailwal discuss AI predictions for 2024, making five bets on topics like AI-generated music, AI podcast segments, AI Pixar shorts, on-device AI models, and open-source multimodal models. They also analyze the New York Times lawsuit against OpenAI, debating fair compensation for content creators and the future of AI licensing.
Main Topics: AI Predictions and Bets for 2024 (Priority: 5/5): Jason and Sandeep make five bets on AI developments by July 1, 2024, including AI-generated Billboard top 100 songs, AI podcast segments, AI Pixar shorts, on-device AI models, and open-source multimodal models displacing ChatGPT. New York Times vs. OpenAI Lawsuit (Priority: 5/5): Discussion on the lawsuit's implications, including fair use, copyright infringement, and the need for licensing models that compensate content creators. AI in Music and Content Creation (Priority: 4/5): Exploration of AI tools like Suno.ai and Midjourney for generating music and video, and the potential for AI to create hit songs or short films. On-Device AI and Edge Computing (Priority: 4/5): Debate on whether Apple or Google will first integrate small AI models into their mobile devices, highlighting Apple's Ferret model and Google's Pixel. Open Source vs. Proprietary AI Models (Priority: 3/5): Comparison of open-source models like LLaVA with proprietary ones like ChatGPT, focusing on reasoning capabilities and multimodal performance. Starlink and Government Contracts (Priority: 2/5): Discussion on Starlink's availability at Costco and the controversy over government subsidies for rural broadband.
Key Arguments: AI-generated music will not reach Billboard top 100 by July 2024 due to music industry control and artist resistance. OpenAI should pay content creators a yearly licensing fee (e.g., $1 per user per year) for training data. On-device AI models will become prevalent, with Apple likely integrating them into iPhones before Google. Open-source multimodal models could surpass ChatGPT in popularity by mid-2024. The New York Times lawsuit could set a precedent for fair compensation in AI training data usage.
Data Points: Midjourney Revenue: $200 million - Midjourney achieved this revenue without venture capital funding. TinyLlama Parameters: 1.1 billion - TinyLlama is a small model trained on 16 A100 chips in 90 days. Starlink Speed: 200 Mbps - Jason reported this speed at his ski house using Starlink. New York Times Subscription Cost: $240 per year - Jason mentions this as the annual subscription price for the NYT. OpenAI Employee Secondary Sale: $8 billion - Potential secondary sale value, compared to NYT's $8 billion market cap.
Pivotal Quotes: "Once you get your first 10 million, you're dangerous. That's what I always tell people: get that first 10 million in the bank." — Jason Calacanis: Reflecting on his sale of Weblogs, Inc. to AOL for $30 million. "My advice to New York Times is to not back down on this one and go to the mat on it." — Jason Calacanis: Advising the NYT to pursue the lawsuit against OpenAI aggressively. "I think the music industry runs like a communist country or any other dictatorship." — Jason Calacanis: Explaining why AI-generated music won't easily break into the Billboard top 100.
Implications: The episode highlights the tension between AI innovation and content creator rights, suggesting that licensing models and legal battles will shape the future of AI. Listeners should expect more on-device AI and open-source models, while the NYT lawsuit could redefine fair use in 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.