This is a wide-ranging interview between creator Nate B. Jones and Chris Best, co-founder and CEO of Substack, recorded to coincide with Substack shipping a Pangram AI-detection integration. Rather than an anti-AI screed, the conversation is a nuanced exploration of what makes online writing valuable when AI can generate plausible text at unlimited scale. Both participants are heavy AI users. The through-line is a search for the line between responsible, intent-driven tool use and the cynical mass production that degrades shared spaces.
What: Best defines slop broadly as "content that nobody believes in," spanning spam, clickbait, and soulless generic material. He was struck by Pangram's finding that roughly 40% of long-form LinkedIn writing is AI-generated.
Why: Substack exists to help people make things they believe in and get paid for it, so slop is its natural foil. The danger is not just bad output. Best frames mass generation as "a denial-of-service attack against the public square." When readers cannot tell what is real, they stop reading new writers and abandon comment sections.
How: Jones adds that the same pattern shows up inside organizations, with docs and Slack messages that carry an "AI smell" and that no one has actually read. The shared premise is that rich human dialogue requires humans to show up and think.
What: Substack's first concrete step is a Pangram scan built into the app. Readers can scan a long-form piece and see an estimate of whether the text went through an LLM. Writers get an optional field to describe their process.
Why: Best is careful that "AI-generated" is not the same as "made without care." The point is not to shut anything down but to add "one tick of transparency" and open a conversation. His theory: responsible AI use should stand up to transparency.
How: The tool surfaces a single, reasonably accurate fact (did this text go through an LLM) and lets readers and writers form their own opinions, giving transparency on both sides of the reader-writer relationship.
What: Jones describes using 20-plus writing skills across Codex and Claude, plus WhisperFlow, to argue with the AI about how ideas are expressed.
Why: LLMs cluster around identifiable "isms": dramatic contrast, over-dramatized headlines, self-talk, and they are cautious by nature. Left alone they water down a bold thesis, so he pushes to preserve clarity and intent.
How: Two distinct modes. When he already has clarity, he dictates a 10-15 minute transcript, then guards against the LLM wrecking that clarity ("you wrecked my clarity, sauce it off"). When he is still finding the idea, he uses Codex as a wall to bounce a tennis ball off, sometimes running 17 drafts in a day to iterate toward conviction. Writing is thinking.
What: Long-form text used to be self-evident proof of work. Now a careful 17-draft piece and a "write me 10,000 plausible takes" dump can both read as AI output.
Why: Best worries this collapses the signal readers relied on. Jones reframes the issue from words to ideas: AI has a "gravitational sink," an averaged central distribution of concepts. His favorite example is that AI always injects ethics and governance into anything about AI, asked for or not.
How: Jones proposes a "Pangram for ideas," a community-built map of the wide distribution of human ideation against the much tighter LLM distribution, measuring concepts rather than text.
What: As models get better, the value of human contribution moves further from the center, not closer. "I don't need you to be a thin wrapper on Claude."
Why: Anything you can easily get from Claude, people will just get from Claude. The remaining value, the alpha, is definitionally at the edges of the distribution, both for ideas and for human connection.
How: Both point to community and conversation as the payoff. Best notes we are a social species; Jones describes the "magical third thing" that emerges in real dialogue. The Odyssey endures because we still argue about it 3,000 years later, and Jones's proudest pieces are the ones that started a conversation.
What: The pair explore how a platform could detect and reward the opposite of slop. Jones's best signal: a coherent thought that acts like a virus, sticking in your head and sparking connections.
Why: Standard attention-seeking algorithms homogenize and sand off edges. The goal is a stronger signal for work that is deeply thought-out, benefiting readers and rewarding creators so they are not drowned out by clickbait.
How: Examples include Paul Graham and Naval, whose ideas change how you see the world, and Anthropic's "J-space" post, which reframed how LLMs compute reality. Crucially, Best argues anti-slop is a pro-AI stance: thoughtful AI leaders publish careful work, not Claude slop.
What: Best's real hope is to spark a community conversation rather than dictate rules, drawing on Substack's spectrum of voices from AI-hostile to frontier-pushing.
Why: The community will help find the new norms that let real work win. The conversation itself can act as an inoculation against slop, because slop is a boring, averaged-out concept.
How: Best notes a platform's stance still matters. He contrasts Substack's transparency move with LinkedIn's "write it for you" button, arguing intent and culture are self-reinforcing.
What: Video currently guarantees effort, but Jones expects the same slop conversation to reach video within a year as generation improves.
Why: New capabilities create norm vacuums. Best coins "Claude-fishing," by analogy to catfishing: the harm is betraying an expectation, not using the tool. He invokes Baumol's cost disease to ask what stays valuable, like the equivalent of a string quartet, as everything else gets cheap.
How: The answer lands on human attention as the last scarce, non-inflationary resource. Best reflects as a parent on modeling how to spend attention well. The closing call is to build a community of conversation about what it means to be anti-slop in the age of AI.
The interview reframes the slop debate away from a binary of human versus AI and toward questions of intent, transparency, and where genuine value lives. Substack's Pangram integration is presented as a first, deliberately modest step: surface one fact, invite writers to explain themselves, and start a conversation. The larger argument is that as AI makes plausible content infinitely cheap, the scarce and defensible things are original ideas at the edges of the distribution and the finite human attention we choose to spend on one another.