Oct 29, 2023 · 30m · another-podcast
LLMs, links, and the death of links
0:00 / 0:00
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this podcast episode, Benedict Evans and Toni-Carran Brown explore the architectural and behavioral evolution of the web, examining how decentralized protocols, closed algorithmic platforms, and generative AI are eroding the traditional role of hyperlinks.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. The hosts hold 83.9% of the talking time here. How this is scored →
The hosts as informed peer 7.7 Guest teaching 1.3 Guest disagreement 0.7 The hosts pushing back 0.9
speaking balance: gold is the hosts, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 8:07 Toni challenges Benedict on creator discovery across platforms
Toni directly rejects Benedict's Instagram comparison by asserting that algorithmic celebrity creation occurs on TikTok, prompting a direct reframe.
Hardest push from the hosts ▶ 24:55 Host rejects non-algorithmic feed idealism as a perpetual motion machineBenedict firmly pushes back against the open-web ideal of uncurated feeds, demonstrating mathematically why chronological ordering fails under high volume.
Biggest teaching moment ▶ 13:56 Toni details how conversational AI terminates external sourcing habitsToni reframes the LLM discussion with concrete user workflow observations, pointing out that AI users cease extracting source links entirely.
The host holds their own ▶ 28:29 Benedict invokes Italo Calvino on feed abundanceBenedict demonstrates deep domain and literary literacy by quoting Italo Calvino's 'Why Read the Classics' to synthesize the information overload dilemma.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | The hosts as informed peer | Guest teaching | Guest disagreement | The hosts pushing back | Why |
|---|---|---|---|---|---|---|
| Defining the Fediverse and the Limits of Interoperability | 8 | 1 | 1 | 1 | Benedict takes control of the conversation, explaining the technical architecture of the Fediverse and contrasting interoperability trade-offs in email and calendars with messaging platforms like Snap and Substack. Toni acts primarily as an interlocutor prompting explanations. | |
| The Shift from Web Links to Algorithmic Discovery | 8 | 2 | 2 | 2 | Benedict provides a detailed historical breakdown of the web's metastructural layers, moving from PageRank and blogrolls to Tumblr and closed recommendation engines like YouTube and TikTok. Toni interjects to clarify that pure recommendation-driven stardom happens on TikTok rather than Instagram. | |
| User Experience, Moderation, and the Open Protocol Dilemma | 8 | 0 | 0 | 0 | Benedict delivers an extended monologue analyzing why open protocol models like the Fediverse struggle with client-side user experience, citing Threads' moderation features like stealth restriction as evidence. | |
| Generative AI, The Death of Links, and Source Extraction | 7 | 4 | 1 | 0 | Benedict discusses how LLMs break the web's link economy by extracting publisher value without referral traffic. Toni educates the host by framing user behavior: unlike web search where users aggregate and clip sources into docs, ChatGPT prompts leave knowledge trapped inside the chat interface. | |
| Personal Curation Habits and the Philosophy of Link Hoarding | 7 | 2 | 1 | 2 | The conversation shifts to personal productivity workflows and link hoarding. Benedict contrasts his newsletter curation pipeline and references Sinofsky's filers versus pilers mental model against Toni's thematic folder structures. | |
| Decentralized Infrastructure, Ad Tracking, and Client Control | 8 | 0 | 0 | 1 | Benedict diagrams historical decentralization from Tim Berners-Lee's vision of local PC servers to modern crypto wallets and Google's Privacy Sandbox, defending the necessity of client control for product sustainability. | |
| Feed Thermodynamics and Calvino's Principle of Sampling | 8 | 0 | 0 | 0 | Benedict applies the laws of thermodynamics to feeds, arguing that uncurated feeds inevitably descend into unusable noise. He concludes by citing Italo Calvino to explain why digital sampling is the only rational approach to information overload. |
Statements from this episode (10)
Evans: Fixed technical standards make subsequent product innovation very difficult
“Once you've got a fixed standard, and everyone has sort of signed up to it, it's very difficult to innovate on top of that.”
Evans: Substack is a publishing platform that disrupted Medium
“Substack has basically massively disrupted Medium. Substack has done what Medium was supposed to do. It's a publishing platform. It's not actually a newsletter platform. It's a blog post by email platform.”
Benedict Evans lacks sympathy for Yelp over Google favoring native maps
“If I search for a restaurant, Google should show me a map. They should just give me a list of links, and Yelp are still very upset about that. Like, I have no sympathy at all”
Evans: RSS Ultimately Failed Because It Was Too Technical
“RSS was way too geeky, and like, there's people who deeply mourn RSS. I used RSS a lot, but it did not work in the end.”
Evans: Fediverse Model Fails Because Client Fragmentation Breaks Social UX
“The problem with the Fediverse model is so much of the experience design has to be in the client. And if it's distributed and there's 50 clients, that doesn't work in the same way.”
Evans: Engagement signal graphs matter more than raw content
“If you're trying to build a competitor to Instagram, I wouldn't want Tony Karen Brown's posts. What I want is who's liked Tony Karen's posts, Tony's posts, and what else did they like? And who liked their posts, and who else liked that? There's like these sort…”
Evans: LLM interfaces resemble Battleship more than document editing
“When you're using LLM, it's a kind of call and response model. You type the prompt, you press go, you wait, And something comes back. I compared it to playing battleships. You know, you call out your move and you find out if you hit anything and then you wait …”
Evans: Open-source clones of commercial products never succeed
“Well, so I was looking at, I looked at Metaverse, sorry, at Fediverse, and saw, this looks like an open source claim of Twitter, and open source claims of commercial products never work. It's the only people who use them are people who care that it's open sour…”
Evans: Ungated, algorithm-free feeds inevitably collapse under content volume
“If you're building a recommendation system that says anybody can publish anything they like, and there's no gatekeeping, and there's no algorithms, guess what? You've got 10,000 things in your feed every day, and you can't find anything. It's like you're tryin…”
Evans: Creator reach declines due to user competition, not algorithm malice
“The other half of it is the reason that you're downranked in the newsfeed is because people don't care, A, people don't care about you that much, B, they followed 50 of your competitors, and B, they have a hundred friends. And there's no solution to that.”