OpenAI and Google race to launch multimodal LLM plus AI demos with Sunny Madra | E1811

Episode Summary

Executive Summary: The episode centers on the rapid shift to AI-powered search and multimodal models, with extensive live demos showing how ChatGPT, Google Search Labs, Canva, and open-source models are changing workflows. The hosts argue that AI is quickly becoming the primary interface for discovery, creation, and productivity, while traditional products risk losing customer relationships, monetization, and relevance if they become mere data sources.

Main Topics: AI as the new interface and aggregator (Priority: 5/5): The conversation opens with concern that ChatGPT could become an apex aggregator, reducing the value of companies that only supply data and no longer own the user interface or relationship. Multimodal AI and future interaction modes (Priority: 5/5): The hosts explain multimodal models as systems that accept and output text, images, audio, video, and code, enabling more natural, flexible interactions beyond plain chat. Google Search Labs vs ChatGPT search (Priority: 4/5): A detailed comparison of Google’s generative search and ChatGPT shows both promise and friction, with Google’s rollout described as powerful but confusingly gated behind settings, accounts, and Chrome. Plugin limitations and vertical software risk (Priority: 5/5): Using Canva as the example, the hosts argue that current ChatGPT plugins are shallow integrations that often underdeliver, while also warning that platforms like Zillow may remove plugins to avoid becoming commodity data providers. Open-source model competition (Priority: 4/5): The discussion highlights Code Llama and Falcon as open-source alternatives narrowing the gap with OpenAI, especially in coding and general language tasks. AI productivity for everyday tasks (Priority: 4/5): The hosts repeatedly show how AI can accelerate party planning, venue research, brainstorming, and content creation by turning messy search tasks into structured tables and editable outputs. AI-generated headshots and bias concerns (Priority: 3/5): The segment on AI headshots covers commercial adoption, cost savings, and a discussion of bias risks in training and reinforcement learning with human feedback.

Key Arguments: Companies that only expose data through plugins lose the interface, customer relationship, and monetization potential. Multimodal systems will become the next major platform shift because users want to interact through text, image, voice, and code in one place. Google may have an advantage in multimodal AI because it controls YouTube, search, and images, but its product rollout is held back by compute and monetization concerns. Current AI plugins are too shallow; the better long-term model is embedding LLMs directly into vertical apps rather than routing users out to ChatGPT. Open-source models are improving quickly and are likely to close the gap with proprietary systems in specific domains like coding. AI is already useful as a daily assistant for brainstorming, planning, and synthesis, but users and organizations need to build the habit of using it consistently. AI-generated outputs can reflect bias from both training data and human feedback processes, not just from model architecture.

Data Points: All In Summit attendance: 1,800 people - The hosts discuss event scale and the difficulty of managing attendance and venue logistics. Party budget: $2.5 million - Jason notes spending this amount on three summit parties. Next-year retreat size proposal: 50 to 100 people - Chamath is described as preferring a smaller retreat-style event rather than a full summit. Theoretical theater capacity range: 4,000 to 6,000 seats - Jason suggests theaters as a scalable venue category for future events. Google search monetization concern: Sponsored links displaced - He notes that generative search answers can reduce clicks on sponsored links. OpenPhone starting price: $13/month - Ad read for OpenPhone business phone service. OpenPhone discount: 20% off first six months - Twist listeners offer. Squarespace discount: 10% off first purchase - Twist promo code offer. Fitbod discount: 25% off - Twist listeners offer for the workout app. AI headshot service price: $29 per person - The hosts mention a market price for AI-generated professional headshots. AI headshot bundle price: $25 for six custom headshots - A competing offer mentioned during the headshot discussion. Code Llama timeframe forecast: By January 2024 - Sunny predicts Code Llama will surpass ChatGPT Code Interpreter capability by then. Falcon model size: 180B parameters - Falcon 180B is cited as a major open-source release.

Pivotal Quotes: "You lose the interface. You lose the customer relationship." — Jason: Arguing why companies should be cautious about relying on AI plugins as their main distribution channel. "I want to leave these AirPods in all day and just be talking to it." — Jason: Describing the ideal hands-free voice-first AI assistant experience. "If your landing page is terrible, I'm out, right?" — Jason: Used as an analogy for why polished, immediate UX matters in AI and web products.

Implications: AI is moving from novelty to core workflow layer. Companies that own interfaces and embedded AI will gain leverage, while data-only providers and shallow plugins risk disintermediation. Users who adopt multimodal, voice, and table-based workflows will work faster and with less friction.