|
|
Was this forwarded to you? Get The Fourth Estate AI Brief in your inbox →
|
|
|
|
View this issue in your browser
Tuesday Brief
The McClatchy Memo and What It Means for the Rest of Us
The interesting thing this week with AI and publishing — McClatchy sent an emoji-laced email celebrating new hires for their 'Content Innovation Lab' less than a week after laying off dozens of journalists. That's the story everyone's talking about. But there's a quieter launch worth your attention too: a new model that can't hallucinate because it literally can't generate text. And Harvard researchers spent a year watching what happens when AI concepts hit real-world production constraints. Three very different signals about where this is all heading.
|
|
MCCLATCHY'S PIVOT
The Rundown: McClatchy laid off journalists at 17 papers last week, then sent staff an email celebrating new hires—including three editors for their AI-powered content operation.
The details:
- More than one-third of Miami Herald's editorial staff was let go; the Lexington Herald-Leader lost more than half its newsroom.
- CEO Tony Hunter told staff in May the company plans to 'agentify the entire enterprise.'
- AI-detection firm Originality found 12-20% of recent Kansas City Star and Miami Herald content was AI-generated.
- The new Content Innovation Lab will focus on lifestyle content: 'barbecue to bourbon, water sports to hiking, and golf to college sports.'
- Job listings for the lab state the 'company's future depends' on embracing AI.
Why it matters for us: This is the cautionary tale for community papers: a chain trading accountability journalism for AI-generated lifestyle content, eroding trust their mastheads spent decades building. If you're a 10,000-circ weekly, your competitive moat is the opposite of this—local reporting that no algorithm can replicate.
Read at san.com →
|
|
|
THE NO-HALLUCINATION MODEL
The Rundown: A former OpenAI researcher launched Jev, a model built for structured decisions rather than text generation—and it can't hallucinate because it literally can't generate strings.
The details:
- Jev outputs structured decisions in 70-500ms, compared to 3-329 seconds for frontier chat models.
- Pricing is $0.042 per million input tokens with free outputs, versus $0.20-$10 per million for traditional LLMs.
- Outputs are constrained to predefined schemas—the model mathematically cannot produce type errors or hallucinations.
- Trade-off: Jev gives up text generation entirely in exchange for speed, cost, and reliability guarantees.
- Target use cases include classification, routing, scoring, and real-time decisions where 100ms response times matter.
Why it matters for us: If your paper routes incoming tips, classifies reader feedback, or scores press releases for newsworthiness, this model could replace brittle if-then rules without the hallucination risk that makes current LLMs dangerous for unsupervised automation. Worth watching as an alternative to prompt engineering for classification tasks.
Read at 1a0bf4f14a877b81 →
|
|
|
THE MOCKUP PROBLEM
The Rundown: Researchers followed senior creatives in TV and advertising for a year and found a recurring problem—AI-generated concepts that violate real-world physics.
The details:
- AI can generate lighting, poses, and compositions that cameras and human actors cannot physically recreate.
- Example cited: a baseball cap casting a face in shadow while eyes remain clearly lit—an impossible lighting scenario.
- Clients approve AI concepts without understanding the production constraints those concepts ignore.
- The speed advantage of AI brainstorming creates new friction points later in the production pipeline.
Why it matters for us: If your ad team uses AI to mock up concepts for local clients, you'll hit this wall: the AI shows something your photographer can't shoot. Better to rough-sketch concepts than over-polish impossible mockups that set expectations you can't meet on deadline day.
Read at hbr.org →
|
|
|
|
Three stories, one throughline: AI is fast, but fast doesn't mean ready. McClatchy is learning that readers notice when content feels hollow. TypeSafe is betting that reliability beats flexibility for automation. And Harvard's researchers are documenting what happens when AI concepts meet real-world constraints. The question for the rest of us is which lessons we learn from watching and which we learn the hard way.
Trevor Slette runs Quadd.ai — AI tools built for publishers.
|
|
|
|
|
|
Run a newspaper. Use Quadd.
AI tools built for small newsrooms. Document intelligence that turns court reports and box scores into copy you can paste into your layout. Audio transcription that finds the quote without you scrubbing for it. AP-style proofreading that tightens copy without flattening voice. Built by someone who knows what a Tuesday night feels like. Free for seven days.
Request a 7-day trial →
|
|
|
