Price:

AI & TECH

Dave Jones trains AI perceptron to purge podcast slop farms

Friday, May 1, 2026 · from 4 podcasts
4 SOURCESPodnews Weekly ReviewThe Pragmatic EngineerNo AgendaPodcasting 2.0
  • An AI researcher is fine-tuning a 500MB filter to identify and block podcast spam farms before they flood indexes.
  • The EU will levy multi-million euro fines on undisclosed synthetic content starting August 2026.
  • AI coding agents break open-source maintenance by flooding projects with unreviewed pull requests.

Dave Jones is waging a technical war against the coming flood. On Podcasting 2.0, he detailed his work on a custom 'LoRA' - a small, 500MB layer of AI weights - trained to spot the specific patterns of podcast spam farms. The goal is a 'perceptron' for the Podcast Index that understands context, not just robotic voices, to block the 550-episode language learning template shows before they overrun the ecosystem.

"Humans detect AI slop instantly - the flat TTS voice, the 55-language template, the lack of credentials - but machines struggle with the nuance."

- Dave Jones, Podcasting 2.0

The slop crisis spans mediums. On The Pragmatic Engineer, Mario Zner described auto-closing every pull request on his Pi coding agent repository. He forces contributors to open a human-written issue first, a desperate bottleneck against the 'valuable garbage' from 'clankers' - AI agents that can fire off thousands of unreviewed PRs. Senior engineer Armen Roner, who interviewed over 30 teams, argues agents lack the human pain feedback loop that prevents complexity, leading to codebases with 'sixteen booleans where only six valid states exist.'

Regulators are now targeting the liability gap. Alberto Betella, CTO of RSS.com, told Podnews that the EU AI Act will mandate disclosure for AI content of public interest starting August 2026, with fines up to 15 million euros or 3% of global turnover. His 'substance test' guides creators on when tagging is necessary, a framework hosts like RSS.com and Spreaker are already implementing for roughly 15% of new episodes.

"The EU AI Act, effective August 2026, mandates disclosure for AI content of public interest, with fines hitting 15 million euros or 3% of global turnover."

- Alberto Betella, Podnews Weekly Review

This isn't just about annoyance - it's about harm. Betella warned that AI slop posing as 'Dr. XYZ' giving medical advice could break ecosystem trust, while Florida's Attorney General opened a criminal probe into OpenAI after a shooter consulted ChatGPT over 200 times for tactical planning. The parallel crises in code and content reveal a core failure: systems built for human-scale contribution are buckling under automated, responsibility-free output.

The response is bifurcating. One path is technical filtration, like Jones's perceptron. The other is architectural reinvention, like Zner's self-modifying Pi agent, designed to be stable and malleable precisely because corporate tools became unreliable. Both acknowledge the same truth: the old gates are broken, and the new ones must be built to understand intent, not just volume.

AI InfrastructureRegulationMedia
ChatGPTPodcast IndexOpenAI

Source Intelligence

- Deep dive into what was said in the episodes

Podnews Weekly Review
Podnews Weekly Review

James Cridland

Fixing podcasting’s AI slop and spam problem: Alberto Betella from RSS.com • May 1

  • James Cridland and Sam Sethi doubt reports of an imminent SiriusXM and iHeartMedia merger, viewing the news as a tactic to prompt shareholder discussions.
  • Global, the UK media group that owns DAX and Captivate, holds almost a third of iHeartMedia. A 2025 FCC rule change allows 100% foreign ownership of US broadcasters.
  • Edison Research data shows radio still dominates US ad-supported audio, claiming 64% of listening time compared to podcasting's 20% and streaming music's 10%.
  • UK audience data from RAJAR reveals live radio accounts for 65% of total audio listening, with podcasts at 10% and services like Spotify and Apple Music at 17%.
  • Alberto Betella of RSS.com defines three categories of AI audio: curated AI, AI spam/infringement, and AI slop - content meant to seem real and often monetized via programmatic ads.
  • RSS.com and Spreaker have implemented voluntary AI disclosure tags in RSS feeds, a step Betella argues builds transparency and helps platforms and advertisers filter content.
  • The EU AI Act, taking effect August 2, 2026, mandates AI disclosure for content of public interest and can levy fines up to 15 million euros or 3% of global turnover.
  • Betella created the 'Substance Test' at shouldidisclose.ai, a framework guiding creators on whether their AI usage is substantial enough to require disclosure under Apple's and regulatory guidelines.
  • Spotify's Q1 2026 report shows 12% year-on-year user growth but a 5% annual and 25% quarterly drop in ad revenue, with auction-based ads now nearing 25% of total ad income.
  • Internal data cited by Cridland shows music makes up 75% of listening time on Spotify, with podcasts at 20% and audiobooks at 5%, challenging perceptions of Spotify's podcast dominance.
  • Libsyn now offers 100GB monthly storage for video files but severely limits audio file storage, a move Cridland criticizes as arbitrary and indicative of a shift toward video and ad platforms.
  • Cridland reports that AI bots constitute roughly a third of all traffic to the Podnews website, highlighting the resource drain of automated scraping on publishers.
Also from this episode: (3)

AI & Tech (1)

  • Betella argues AI slop is dangerous for sensitive topics like health, where wrong AI-generated advice can cause real harm, while being merely annoying for generic categories.

Protocol (1)

  • Sam Sethi expresses skepticism about Bitcoin-based podcast micropayments, suggesting stablecoins integrated with traditional payment rails like Visa are a more viable path for mass adoption.

Stablecoins (1)

  • Stablecoins processed a$7 billion annualized run rate, growing 50%, as companies like Meta use them for creator payouts in markets like Colombia and the Philippines.
RegulationModelsBig TechEnterpriseAdoption
The Pragmatic Engineer
The Pragmatic Engineer

The Pragmatic Engineer

Building Pi, and what makes self-modifying software so fascinating • Apr 29

  • Mario Zner auto-closes all first-time pull requests to filter out AI-generated spam. His GitHub workflow posts a comment asking for a human-written issue; agents ignore the comment, but humans respond, earning future PR privileges.
  • Armen Roner sees a future reckoning where engineering teams realize they cannot maintain their codebases without AI providers, creating dangerous vendor lock-in. He expects this dependency and its cost to become a major industry conversation.
Also from this episode: (8)

AI & Tech (6)

  • Mario Zner built Pi because he wanted a simple, stable agent after Claude Code became unreliable. He reverse-engineered Claude Code and found its system prompts and tool definitions changed with every release, breaking his workflows.
  • Pi is a minimalist, self-modifiable coding agent. Its core provides read, write, edit, and bash tools with extensive hooks, allowing users to ask Pi to modify its own TUI, add features like MCP support, or tailor it for specific workflows like game development.
  • Armen Roner interviewed over 30 engineering teams and found AI agent adoption exploded after holiday breaks like Christmas 2024. He says adoption requires a two-to-three week learning period that is difficult during normal work sprints.
  • Armen Roner argues AI-generated code lacks a human's pain feedback loop. Senior engineers say no to avoid future complexity pain, but agents and junior engineers empowered by agents say yes, accelerating codebase bloat and deterioration.
  • Non-engineers like product managers now directly submit AI-generated pull requests. Armen Roner cites cases where marketing teams modify websites and sales teams build non-existent features into demos that land in repositories.
  • Both hosts argue the real value of AI agents is automating tedious work to free up human time for design and polish, not maximizing token output. They say the current hype pushes for unsustainable speed at the cost of quality and engineer well-being.

Coding (2)

  • Mario Zner believes MCP is overly complex and non-composable for developer tasks, favoring CLI-like code execution. He argues agents are creative with CLI pipes but MCP servers that dump entire API specs create useless tool sprawl.
  • Armen Roner warns the industry's 'dark factory' approach of deploying armies of agents with vague specs will produce low-quality software. The output quality is bounded by the mediocre training data the models use to fill specification gaps.
AgentsOpen SourceCodingStartups
No Agenda Show
No Agenda Show

Adam Curry

1863 - "Nekkidly" • Apr 26

  • Adam Curry and John C. DeVorex hosted "No Agenda" Episode 1863 on Sunday, April 26, 2026. John C. DeVorex noted widespread "false flag" claims regarding an unspecified event.
  • During an interview about a reported false flag at the White House Correspondents' Dinner, Fox News allegedly cut off Aisha Hasni as she was about to reveal critical information.
  • Over 200 journalists signed a letter demanding that Donald Trump be challenged on press freedom at the White House Correspondents' Dinner, which also featured a mentalist instead of a comedian during his last attendance.
  • The shooter's LinkedIn manifesto targeted "pedophile rapist and traitor" Trump administration officials, specifically excluding a "Mr. Patel." His brother had previously alerted local police to alarming writings.
  • Alex Jones claimed "globalist mad scientists" created an "intergalactic communication system," a term J.C.R. Licklider used in the 1960s to envision the internet as a nuclear-attack-resilient, distributed network.
  • John Stossel's 2017 report on the SPLC criticized its practice of labeling critics of radical Islam as "anti-Muslim extremists" and highlighted its growing endowment, then over $320 million.
  • Chris Cuomo defended the SPLC, noting its historical cooperation with federal law enforcement against hate groups, a relationship he claimed the Justice Department recently terminated.
  • John C. DeVorex is optimistic Apple's integrated chips and universal memory in devices like the Mac Mini and Mac Studio position them well for local AI model inference, unlike competitors who cram phones with "AI garbage."
  • Florida's Attorney General, James Uthmeyer, opened a criminal investigation into OpenAI after an FSU shooter allegedly consulted ChatGPT over 200 times for planning advice.
  • A former Pfizer Europe chief toxicologist testified in Germany that the Comirnaty vaccine's carcinogenicity and reproductive effects were not adequately tested before fast-track approval.
  • Pfizer's post-marketing report noted over 1,200 suspected deaths within two months of Comirnaty's approval; a Paul Ehrlich Institute report identified 2,133, suggesting an actual 60,000 deaths in Germany with a 30x underreporting factor.
  • The Pfizer toxicologist stated that Comirnaty was not tested for preventing severe illness or death, invalidating the courts' assumption of a "positive risk-benefit ratio." Mortality in Germany rose significantly from 2021 to 2022.
  • Dr. Eric Berg highlighted that a 2007 law mandating drug study results be posted, with a $13,000 daily fine for non-compliance, has led to zero FDA fines in 19 years, totaling $19 billion owed by pharma.
  • King Charles III and Queen Camilla will visit the U.S. to commemorate 9/11 and America's 250th birthday, including the Yorktown battlefield, a symbolic location for British defeat.
  • British commentators viewed King Charles's U.S. visit as an "embarrassment" due to Donald Trump's past insults towards British troops, NATO, and the Royal Navy, despite its purpose as a "soft power" diplomatic effort.
  • Manosphere podcasters are turning on Donald Trump, criticizing his unfulfilled promises on deportations, Epstein files, and gasoline prices, a shift CNN and MSNBC suggest could undermine his public image.
  • Adam Curry emphasized that "No Agenda" provides analysis, not support, aiming to offer alternative perspectives by questioning mainstream narratives, a strategy he believes strengthens listener's beliefs or prompts questioning.
  • A Michael Jackson biopic, "Michael," starring his nephew Jafar Jackson, explores his rise from the Jackson 5 and escape from his father Joe Jackson's control, concluding at his 1984 *Thriller* superstardom peak.
Also from this episode: (7)

Politics (2)

  • Margaret Brennan linked an alleged shooting at the dinner to the Second Amendment, citing 564 threats against judges and nearly 15,000 against lawmakers last year.
  • Dame Rhonda described how an SPLC lawsuit, *Ricky Wyatt v. Alabama Department of Mental Health*, led to such high standards that Alabama and other states defunded mental health care.

Corruption (1)

  • A 31-year-old alleged shooter, identified as Allen, traveled by train from Southern California with multiple weapons, including a shotgun, handgun, and knives, and shot a Secret Service officer in body armor.

Business (1)

  • John C. DeVorex asserted that Enron, during its bandwidth trading, undermined the internet's original peering system by introducing charges, contributing to its eventual centralization.

AI & Tech (1)

  • Anthropic has substantially increased Claude AI service costs, with monthly subscriptions reaching $200 and additional credits costing $2 every 30 seconds of usage, suggesting an IPO strategy.

War (2)

  • A leaked Pentagon memo reportedly considered sanctions against NATO allies, including reviewing Britain's ownership of the Falklands, for not supporting the U.S. in the Iran war.
  • Argentina is rearming with F-16 jets from Denmark, supported by U.S. missiles, raising concerns for the UK's ability to defend the Falklands, given its limited military footprint there.
SafetyRegulationMacroSociety
Podcasting 2.0
Podcasting 2.0

Adam Curry

Episode 258: Perceptron • Apr 24

  • Alex Sanfilippo hosted a town hall addressing podcast guest spam, where Tom Rossi proposed removing emails from RSS feeds. Adam Curry suggested a "booking tag" for RSS feeds, a solution Alex and Daniel J. Lewis are developing.
  • Dave criticizes Apple and Spotify for their opaque, "editorial team"-driven podcast curation, contrasting it with most other podcast apps that avoid subjective recommendations. He argues app developers should offer editorial to "delight users."
  • Paul from Godcaster reports App Store approvals are taking longer, especially for "wrapper" apps; his last Android app release took over a week. Dave recounts Apple rejecting his functional recruiting app as a "promotional advertisement," despite its utility.
  • DeepSeek V4 Pro boasts 1.6 trillion parameters and a 1 million token context window, signaling significant AI advancement. Apple's internal silicon and universal memory position it well for AI integration, despite current restraint on "AI nonsense."
  • Dave utilizes Together.ai for stable GPU rental to process and summarize long audio efficiently at about 15 cents per hour. The Podcast Index blocks data center traffic with heuristics, preventing "slopocalypse" from bot armies using distributed proxies.
  • Adam Curry defines "slop" as repetitive, low-quality AI-generated content, comparable to "pig feed." He argues humans can identify AI slop intuitively, even if they cannot articulate the precise definition.
  • He highlights a paradox: while words are fungible conduits for meaning, their specific choice is critical, especially in propaganda, which aims to drive specific actions.
  • Dave developed an agent identifying AI slop videos using red flags like generic phrasing, no human presence, and monotone TTS narration. This agent flagged 7 AI slop clips from 55 language courses on Spreaker, detecting two TTS voices across all languages.
  • Dave states that 90% of AI model training effort dedicates to preparing high-quality, diverse training data to prevent memorization and achieve generalized learning. He is building a problematic feed database of 25,000 good podcasts for this purpose.
  • Dave explains Low-Rank Adaptation (LoRA) as a method to fine-tune large language models by adding small (under 500 MB) custom weight adapters to a base model. This approach allows for highly customized outputs without retraining the entire model, enabling rapid updates.
  • Adam Curry notes public domain LibriVox recordings are being reposted for ad revenue, sometimes with poor human narration. He suggests a high-quality AI narrator could improve these, leading to a need for individual "perceptrons" to filter content.
  • Adam Curry hit his token limit on GitHub Copilot's $100 plan, attributing it to a suspected default model change to Opus 4.7 ('extra high effort'). Dave terms this 'token inflation,' a way to effectively raise costs through increased token consumption per request.
  • Dave recommends running local models like Triquin 3.6 (35B A3B model) on Open Code, praising its lightning-fast inference speed and immediate output.
Also from this episode: (5)

Lightning (1)

  • Adam Curry's Lightning node was not live in the splits, preventing him from receiving live boostagrams on the previous show until Eric PP identified the issue.

History (1)

  • Adam Curry recalls Steve Jobs' initial vision for the iPod Touch and iPhone centered on web apps, with a proprietary App Store becoming a necessity only after unforeseen Wi-Fi issues.

BTC Markets (1)

  • Adam Curry received "boost spam" (one Satoshi from "Satogram") after activating his node, but Eric PP noted Helipad software includes a feature to filter boost amounts below a user-defined threshold.

Business (1)

  • Adam Curry downplays Tim Cook's reputation as a business genius, attributing Apple's market cap growth partly to stock buybacks. Dave acknowledges Cook's positive impact on supply chain management.

Psychology (1)

  • Dave discusses Noam Chomsky's theory of universal grammar, suggesting language is an innate, wired-in human faculty rather than merely a learned skill.
ModelsEnterpriseStartupsRegulationMedia

Related Stories