Plausibly Deniable
← Episode archive

AI News And The Manufactured Internet | Brian Chau

About this episode

Brian Chau has figured out how to use AI for journalism without producing AI slop: make the models read mountains of documents, then make humans write the story.

Brian is the founder of Effort News, a publication using long-context AI to comb through federal databases, financial records, legal filings, and other giant piles of public data for stories everyone else missed. He joins Lukas and Saila to explain the refugee-resettlement grant investigation that convinced him to launch Effort, why ordinary journalism is often built around stories sources want “placed,” and what changes when you can query an entire government dataset instead.

They also get into why frontier models are becoming better at math and worse at prose; Claude’s sycophancy; the different sources models inherently trust; taxpayer-funded NGOs; the funding behind anti-data-center and anti-AI activism; SBF’s Anthropic investment; Nick Shirley’s fraud counters; the EU’s attack on the open internet; zoning, NIMBYs, giant trucks, and “full-force centrism”; writing separate websites for humans and AIs; why Brian’s p(doom) is negative; and whether Effort can become a new source of truth for the AI age.

Chapters

  1. 00:00 “Unaligned NGOs” and the anti-AI money trail
  2. 01:06 Welcome and Brian Chau
  3. 02:01 From machine learning to Effort News
  4. 02:47 Why AI writing keeps getting worse
  5. 08:21 Sycophancy and models that instantly agree with you
  6. 09:37 Which sources different models trust
  7. 13:54 How Effort News finds stories
  8. 15:34 The refugee-resettlement grant investigation
  9. 25:00 Data-first journalism versus planted stories
  10. 27:08 A million AI agents versus government databases
  11. 28:58 Data centers, unions, and political extortion
  12. 31:29 Why Brian is optimistic about the next 20 years
  13. 33:19 Europe, censorship, and the open internet
  14. 37:30 A white pill wrapped in black pills
  15. 42:22 Taxpayer-funded media and manufactured consensus
  16. 44:10 What does it mean for discourse to be “fake”?
  17. 44:57 Who funds anti-data-center activism?
  18. 50:13 Nick Shirley, fraud counters, and due diligence
  19. 57:49 Full-force centrism
  20. 58:28 Zoning, NIMBYs, and the affordability crisis
  21. 1:04:04 “Things can just be better”
  22. 1:06:43 How regulation made trucks enormous
  23. 1:09:03 Rebuilding a shared political reality
  24. 1:13:02 Tucker, Candace, and the right-wing influencer problem
  25. 1:15:13 How Effort News is funded
  26. 1:16:00 Who gets to shape AI’s values?
  27. 1:17:09 Writing one website for humans and another for AIs
  28. 1:20:16 How much of culture is manufactured?
  29. 1:22:50 The anti-AI funding network
  30. 1:24:47 SBF’s attempted crypto regulatory monopoly
  31. 1:27:06 Why Brian’s p(doom) is negative
  32. 1:29:01 The UK, France, and nuclear instability
  33. 1:31:01 The accidental founding of Effort News
  34. 1:32:25 Subscribers, B2B data, and a future newswire
  35. 1:35:24 Building a new source of truth
  36. 1:39:29 The Plausibly Deniable exclusive
  37. 1:42:48 Publicists, the Vice mafia, and founder-led media
  38. 1:47:13 Effort’s team and the disgusting CTA
  39. 1:48:13 Pre-made Twitter threads and “we read AI text”
  40. 1:49:36 Why real scoops sit next to total slop
  41. 1:54:24 Substack, data visualization, and $80,000 subscriptions
  42. 2:02:23 Final plugs and farewell

Keep exploring

Try the personality tools · Get the next episode by email