If the AI Writes Slop, You Gave It Slop
Morning Brew starter Alex Lieberman built a bot that drafts only from his own spoken answers, and says thin output exposes thin thinking.
How I AI · The AI content machine that turns ideas into posts that don't sound like slop | Alex LiebermanAlex Lieberman, who started Morning Brew in college and now works at 10X, an applied AI company that helps midsize and large firms adopt AI, was showing the interviewer, Claire Vale, a product leader, how his homemade posting machine drafts LinkedIn posts.
Asked how he avoids slop when AI writes so badly out of the box, Lieberman said the outcome comes down to the interview step. In his view, the only time the machine produces slop, it indicts the person for not sharing good enough ideas, not the model for writing badly.
What he means, he argues, is that the system only reveals the quality of the thinking put into it. The first step, which he calls the Oracle, scans the last seven days of his messages and reading and ranks 15 possible ideas using its scoring system for a good spike. Then an interview panel questions him to pull out details, and he answers by talking.
The draft that follows, he says, draws largely on that interview transcript, apart from the opening and ending, which he may add separately. The job of the writer, in his phrase, is "shaping the clay of the content" rather than inventing new material. During the demo, he flagged one of his own draft lines, about two smart people disagreeing over the hottest job in AI, as feeling cringey and generic. He said he would label it that way and add the note to his lessons file, and he recently added a lesson telling the system to offer him more hooks to compare.
Vale replied, "Uh-oh, so what you're saying is we are actually the slop." Lieberman said, "AI slop is hilariously people just pointing the finger at themselves and saying 'I'm not intelligent enough.'"
Questions people ask
How does his transcript-only rule actually work?
He answers interview questions by voice, and those answers become a long transcript. The system is instructed to build the draft largely from that transcript, with only the opening and closing possibly added, so the post reflects his stories and opinions.
If ideas matter more than writing, where does he get them?
Lieberman says finding a fresh idea is the hardest part of posting. His Oracle is meant to solve blank page paralysis by surfacing 15 ranked ideas from the past week, so he starts from something concrete instead of a blank box.
Does that mean AI cannot help a beginner with nothing to say?
In his view, probably not much. He argues the interview must supply stories, views, and specifics. Without that material, he says, the draft will be thin no matter how polished the sentences sound.