Episode Summary
Executive Summary: The episode centers on Meta’s surprising rise in open-source AI, especially Llama 3, and debates whether Facebook/Meta can be trusted as a business partner versus as infrastructure. The hosts compare open vs. closed models, assess Grok’s real-time X-based search, discuss AI’s military/autonomous driving implications, and demo two startups: a prompt-to-deployed-app builder and an AI vertical search engine.
Main Topics: Meta/Llama 3 and open-source AI (Priority: 5/5): The hosts react to Llama 3’s strong benchmark performance, arguing that Meta has become a major force in AI by releasing smaller, highly capable open models that can run privately and at scale. Can Facebook be trusted? Infrastructure vs. partnership (Priority: 5/5): A long-running debate contrasts Meta’s unreliability as a business partner with its strong track record in infrastructure open source (Open Compute, React, PyTorch). Grok as real-time research over X (Priority: 4/5): They demo Grok’s ability to summarize live social discourse and cite sources, arguing X’s noisy but timely data can be mined for valuable signal if ranking/filtering improves. AI in defense and autonomy (Priority: 4/5): They briefly discuss the U.S. Air Force’s AI F-16 dogfight and tie it to progress in autonomous driving, including recent improvements in Tesla FSD and expectations for robotaxi deployment. Startup demos: prompt-to-app and AI vertical search (Priority: 4/5): Two products are evaluated: a system that generates and deploys full apps from prompts, and a vertical search engine that synthesizes results from Amazon, YouTube, and Reddit into answer pages. Design, execution, and startup advice (Priority: 3/5): The hosts give practical feedback on product design, hiring designers cheaply, and using prior companies/failed attempts as execution lessons for founders.
Key Arguments: Meta’s Llama 3 shows that open source can outperform proprietary incumbents at smaller scale and with better deployment economics. A company can be trusted for infrastructure contributions if it has a consistent track record, but not as a business partner if it controls APIs or platform access. Open source wins enterprise trust because companies want control, private deployment, and less vendor lock-in; commercial vendors still win on support. X/Twitter has a unique advantage for real-time information because it contains immediate human reactions, but the openness and anonymity create noise that LLMs must filter. Grok is useful because it can summarize current discourse and provide citations from the platform itself, though some answers still reflect bias or imperfect sourcing. Meta’s open compute, React, and PyTorch demonstrate that giving away infrastructure can improve ecosystem supply chains and ultimately strengthen Meta’s own position. Prompt-to-deployed-app tools represent a new layer above copilots and agents: an end-to-end “consulting firm” for software creation. Vertical AI search can modernize the old Mahalo model by dynamically aggregating sources instead of relying on manually edited pages.
Data Points: Llama 3 model size: 8 billion parameters - The smaller Llama 3 model is described as outperforming the original GPT-4 in benchmark comparisons. LLM deployment footprint: Relatively small hardware footprint - The discussion emphasizes that smaller open models can run at scale and inside enterprise firewalls. Enterprise preference for open source: 82% - They cite an a16z survey saying most enterprises prefer open-source AI over closed source. LinkedIn user base: 1+ billion users - Used in an ad read to explain why LinkedIn Ads are effective for B2B marketing. Senior executives on LinkedIn: 180 million - Ad read highlights the density of decision-makers on LinkedIn. C-suite executives on LinkedIn: 10 million - Used to illustrate access to high-value business buyers. LinkedIn Ads ROI: 2 to 5 times greater - Claimed return compared with other social platforms in the ad copy. Zendesk startup program: 6 months free - Promotion offered to qualifying startups. Openly customer service response time: 30 seconds or less - Example of a Zendesk customer handling chats and calls very quickly. Openly customer satisfaction: 96% - Reported as a result of their Zendesk-powered support experience. Openly funding: $100 million - Mentioned as a recent raise for the Zendesk customer example. AI strategy roadmap price: $5,000 - Curotech offers a tailored AI strategy roadmap at 50% off the normal price. Curatech first-month discount: 20% off - Ad copy for startup engineering hiring support in Latin America. Tesla FSD 12 observations: About 1 intervention every 10 minutes - Host reports on street driving performance; highway driving was described as bulletproof. Rides on FSD 12: 2 drives - Personal anecdotal testing of the software. X post citations example views: 44 views - One cited Trump-related post had very low engagement, prompting discussion of source quality. Met a 10,000+ pitch milestone: 10,000 in-person pitches - Host uses this to explain why he recognizes repeated startup ideas and patterns. Minimum designers used in a hiring strategy: 3 to 5 designers - Advice for getting multiple design options for around $5,000 total. Designer compensation example: $1,000 each - Proposed budget for quick design bake-offs. Meta stock price: $484 - Referenced while discussing market reaction and earnings expectations. Meta stock move: +0.77% - Short-term movement mentioned during the stock discussion.
Pivotal Quotes: "As a business partner, no, I don't believe you can trust Facebook." — Jason Calacanis: Directly states caution about Meta in business/partnership settings. "With them as a partner, as a business partner, they will slit your throat." — Jason Calacanis: Hyperbolic warning about Meta’s history of changing platform terms and rug-pulling partners. "The model at the top level like meta.ai... What it did was aggregated the large corpus of data and giving you the average of what it sees there." — Jason Calacanis: Explains how Grok can summarize X’s real-time signal and noise.
Implications: Open-source AI and real-time social data are becoming strategic infrastructure. Meta may dominate by giving away core tech, while products that can filter noisy live data or auto-build software could reshape startups, search, and enterprise workflows.
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.