Episode Summary
Executive Summary: In this AI-focused episode, Jason Calacanis and Sundeep Madra analyze Google's latest AI updates, including Notebook LM and Gemini Pro, while discussing the commoditization of large language models. They critique Google's branding strategy, praise Gemini's benchmark performance, and explore the implications of open-source models like Mistral 8x7B. The conversation also covers AI-generated influencer images and tools for upscaling and aging photos, highlighting the transformative impact on marketing and content creation.
Main Topics: Google's AI Updates and Branding Critique (Priority: 5/5): Discussion of Google's new products (Notebook LM, Gemini Pro) and criticism of their confusing branding strategy with multiple names like DeepMind, Bard, and Gemini. Commoditization of Large Language Models (Priority: 5/5): Analysis of how models like Mistral 8x7B are commoditizing GPT-3.5 level capabilities, making advanced AI accessible to smaller players. AI-Generated Influencer Images and Marketing Collateral (Priority: 4/5): Demonstration of creating realistic AI-generated influencer photos using stable diffusion models, with implications for e-commerce and marketing. Benchmark Performance of Gemini vs GPT-4 (Priority: 4/5): Comparison of Gemini Ultra and Pro against GPT-4 on various benchmarks like MMLU and GSM8K, showing Google's competitive position. Poe Platform and Multi-Model Access (Priority: 3/5): Overview of Poe by Quora as a subscription service providing access to multiple AI models, including GPT-4, Claude, and Mistral. AI in Education and Test Scoring (Priority: 3/5): Discussion of Gemini's ability to score tests, identify mistakes, and provide tutoring, potentially transforming education. Photo Upscaling and Aging Tools (Priority: 2/5): Demonstration of tools that upscale and age photos, making AI-generated images more realistic and useful for marketing.
Key Arguments: Google's branding is a weakness; they should follow YouTube's model with independent brands like 'Bard' instead of 'Google Bard'. GPT-3.5 level models have been commoditized by open-source models like Mistral 8x7B, reducing the value of large proprietary models. AI-generated influencer images can replace expensive photo shoots, saving startups millions in marketing collateral costs. Gemini Ultra outperforms GPT-4 on several benchmarks, proving OpenAI's lead is not insurmountable. Attribution in AI is possible and important for monetization and content creator relationships. The combination of AI image generation and upscaling tools can produce realistic marketing materials without human models.
Data Points: Gemini Ultra MMLU Score: 90% - Multiple choice questions in 57 subjects, outperforming GPT-4's 87%. Gemini Ultra GSM8K Score: 94.4% - Grade school math benchmark, compared to GPT-4's 92%. Mistral 8x7B Model Size: 8x7B parameters - Mixture of experts model that can outperform larger models like Llama 2 70B. Model File Size: 6.62 GB - Size of the Juggernaut XL model used for AI image generation. Poe Subscription Cost: $199 per year - Cost for access to multiple AI models including GPT-4, Claude, and Mistral. Jason's Ski Speed Record: 40.2 mph - Personal best speed, compared to Sundeep's 54 mph.
Pivotal Quotes: "GPT-3.5 level models have been commoditized. There's nothing special about them." — Sundeep Madra: Discussing the impact of open-source models like Mistral 8x7B on the AI landscape. "A large language model on the internet is worth a billion dollars if it's one of the top five. It's not worth a hundred billion." — Jason Calacanis: Arguing that the value of LLMs will decrease as they become commoditized. "Google's strength is in the underlying technology, their weakness is in branding, and often in UX." — Sundeep Madra: Critiquing Google's approach to product naming and user experience.
Implications: The commoditization of LLMs will democratize AI, enabling startups to compete with giants. AI-generated marketing collateral will disrupt traditional advertising and e-commerce. Google's improved AI performance signals a competitive market, but branding and UX remain critical for consumer adoption.
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.