Episode Summary
Executive Summary: The episode argues that ChatGPT’s rapid rise and declining web traffic signal a structural shift: AI chatbots are becoming the primary interface for information while extracting far more value from publishers than they return. The hosts connect this to Grok’s prompt-driven ideological leakage, Meta’s model delays, signs of scaling limits across major labs, and the emergence of agentic commerce via Perplexity and PayPal.
Main Topics: ChatGPT as the “last website” (Priority: 5/5): SimilarWeb data shows ChatGPT climbing to the world’s fifth-most visited site while major incumbents decline, fueling the idea that AI chatbots may become the dominant destination for information consumption. AI vs. publisher economics (Priority: 5/5): Cloudflare’s crawl-to-click ratios illustrate how search and AI systems increasingly reuse web content without sending meaningful traffic back, threatening the ad/subscription model that sustained the web. Grok’s system prompt and hidden ideology (Priority: 5/5): An unauthorized prompt change caused Grok to inject white genocide propaganda into unrelated chats, highlighting how chatbot behavior is shaped by hidden instructions and could subtly steer users’ beliefs. Scaling limits and Meta’s Behemoth delay (Priority: 4/5): Meta’s delayed flagship model is framed alongside OpenAI and Anthropic delays as evidence that bigger models are not delivering expected leaps, challenging the assumption that scale alone solves model quality. Enterprise ROI and implementation gap (Priority: 4/5): The discussion separates model capability from business deployment, arguing that many AI disappointments come from poor implementation and mismatched expectations rather than failure of the technology itself. New distribution models: shopping and structured content (Priority: 4/5): Perplexity’s PayPal integration and MasterCard’s agentic tokens show how AI systems may absorb commerce and structured information directly, bypassing websites as the default transaction layer. Open source as an alignment answer (Priority: 3/5): The hosts briefly discuss the idea that open source could reduce the risk of hidden control over AI systems by making model behavior more transparent and user-aligned.
Key Arguments: ChatGPT’s rise to a top-five global website suggests AI may be displacing the open web as the main entry point for information. The economic bargain of the web is breaking: AI companies crawl and ingest content at far higher rates than they send traffic back. Hidden system prompts matter because they can encode ideology or behavior into chatbots without users realizing it. Grok’s white genocide incident is presented as an extreme but revealing example of prompt-driven model behavior gone wrong. Meta, OpenAI, and Anthropic all appear to be hitting diminishing returns from scaling larger models. Many enterprise AI disappointments are likely due to poor implementation, unrealistic expectations, and insufficient context handling rather than total model failure. The future may involve structured data and direct feeds into AI systems instead of human-readable web pages. Agentic shopping and payments could become a major consumer use case as AI interfaces absorb more of the commerce flow.
Data Points: ChatGPT web rank: #5 globally - SimilarWeb ranking cited in the discussion, behind Google, YouTube, Facebook, and Instagram ChatGPT month-over-month traffic change: +13% - Desktop traffic growth while major platforms were declining Google crawl-to-visitor ratio 10 years ago: 2:1 - Cloudflare CEO Matthew Prince’s historical comparison Google crawl-to-visitor ratio 6 months ago: 6:1 - Cloudflare data showing worsening traffic return to publishers Google crawl-to-visitor ratio recently: 15:1 - Cloudflare data showing a sharp increase in crawls per visitor OpenAI crawl-to-visitor ratio: 250:1 - Cloudflare estimate cited for OpenAI Anthropic crawl-to-visitor ratio: 6,000:1 - Cloudflare estimate cited for Anthropic OpenAI answer rate on Google: 75% - Prince says Google now answers 75% of queries on Google without sending users to source sites World History Encyclopedia traffic hit: 25% - CEO cited AI Overviews causing a quarter drop in traffic IBM CEO survey sample: 2,000 CEOs - Global survey on AI adoption and ROI CEOs actively adopting AI agents: 61% - IBM survey result AI initiatives delivering expected ROI: 25% - IBM survey result AI initiatives scaled enterprise-wide: 16% - IBM survey result CEOs saying fear of falling behind drove investment: 64% - IBM survey result CEOs expecting investments to pay off: 85% by 2027 - IBM survey result Workforce needing retraining/reskilling: about one-third - IBM survey result CEOs hiring AI-related roles not existing a year ago: 54% - IBM survey result Cohere revenue (Reuters): $100M annualized in May 2025 - Positive revenue marker reported by Reuters Cohere prior investor target: $450M ARR by 2024 - Information report cited as an overambitious expectation Cohere revenue (Information): $70M in Feb. 2025 - Alternative figure showing slower growth than expected AlphaEvolve problem set: 50+ open problems - DeepMind said the system was applied across math and CS domains AlphaEvolve rediscovered state-of-the-art solutions: ~75% of cases - DeepMind claim about benchmark performance AlphaEvolve improved best-known solutions: 20% of cases - DeepMind claim on open problems Gemini training time reduction: 1% - AlphaEvolve helped optimize Gemini’s training Gemini kernel speedup: 23% - DeepMind claim about architecture optimization
Pivotal Quotes: "ChatGPT, the last website." — Host: Used to frame the idea that generative AI may become the primary destination for information while the rest of the web declines "For every two pages Google crawled, they sent you one visitor... Now it's up to 15 to one." — Matthew Prince (quoted by host): Cloudflare’s explanation of how crawl-heavy systems are extracting more value from publishers than they return "The real alignment problem is who controls the AI." — Daniel Jeffries (quoted by host): Used to argue that open source could reduce hidden control over model behavior
Implications: If AI chatbots become the main interface to information, publishers lose traffic and influence while model makers gain gatekeeping power. Expect more fights over content rights, prompt transparency, enterprise ROI, and commerce integration inside chat.
About Big Technology Podcast
The Big Technology Podcast takes you behind the scenes in the tech world featuring interviews with plugged-in insiders and outside agitators. Alex Kantrowitz, a Silicon Valley journalist who's interviewed the world's top tech CEOs — from Mark Zuckerberg to Larry Ellison — is the host.