A practical breakdown of using Jev efficiently, starting with the costly mistake of running it after expensive AI work. Explore smarter filtering, precise evaluation criteria, Grill Me workflows, pre- and post-work checks, and a reliable path for reviewing content. The discussion also covers evidence inspection, human oversight, and scaling these safeguards across teams.
08:07
forum Ask episode
web_stories AI Snips
view_agenda Chapters
auto_awesome Transcript
info_circle Episode notes
question_answer ANECDOTE
Eric Siu Learned Jev Was Running Too Late
Eric Siu discovered he was running Jev after expensive models had already completed the work.
A revised workflow filtered 189 of 517 chunks before costly processing, though that result measured routing, not guaranteed savings.
volunteer_activism ADVICE
Define Jev Gates Before Processing Content
Define precise gates before using Jev, such as non-AI content, originality, legitimate statistics, recent citations, and meaningful improvement.
Choose whether each criterion requires true-or-false, multiple-choice, or ordered scoring evaluation.
volunteer_activism ADVICE
Use Grill Me To Discover Better Criteria
Use Matt Pocock’s Grill Me skill to uncover evaluation criteria through sequential questions.
One-at-a-time questioning works best when each answer should shape the next question and you have enough time.
Get the Snipd Podcast app to discover more snips from this episode
👉 Single Brain - Revenue agents to scale faster using our unified intelligence layer. https://www.singlebrain.com
I was using Jev after the expensive model had already done the work. In this video, I show the mistake, the filtering workflow I changed, and how I define criteria before trusting the output.
We walk through real content examples, Grill Me versus Grill Me Overview, pre- and post-work checks, and why a human still needs to inspect the evidence. In one batch, Jev passed 189 of 517 chunks to the costly model. That is a filtering result, not a measured percentage of money saved.
Clarifications: The 198/225 and 189/517 figures come from different datasets, not a controlled before-and-after test. Jev’s Noul returns the probability of yes; Choice selects an option; Score rates against ordered criteria. Score is not a confidence score.
Chapters:
00:00 The biggest Jev mistake
00:14 Filter before the expensive model
01:07 Define the right criteria
02:23 Grill Me: questions one at a time
02:43 Work with Single Grain
03:05 Grill Me Overview: batch the questions
04:36 Add pre- and post-work gates
05:15 The real article and evaluation path
06:08 Keep a human in the loop
06:52 Teach your team the workflow
Footage and sources: Eric Siu’s original camera and screen recording. Single Grain website footage is retained from the original edit. Grill Me is credited in the narration to Matt Pocock; the article and workflow excerpts appear within Eric’s recorded screen demonstration.
How to Connect
IG: https://www.instagram.com/ericosiu/
X: https://x.com/ericosiu
👉 Single Grain - AI marketing firm https://www.singlegrain.com
👉 Get 25k Words of AI-Powered SEO Content: https://clickflow.com/
👉 Growth Newsletter for top marketers: https://levelingup.beehiiv.com/subscribe
👉 Connect your tools with Single Brain Gateway: https://www.singlebrain.com/gateway
Leveling Up with Eric Siu