← sources

How do you keep up with AI when new things happen every day? I used JEV to help me in this: (used ChatGPT for structure only)

source: reddit (reddit:AI_Agents)published 2026-10-02 13:35ingested 2026-10-03 02:35

retained text

Here's what I've explored so far.

**1. Near-duplicate detection**

Different sources often publish the same story. Using LLMs to compare every article would increase API costs, so I explored **MinHash + Locality-Sensitive Hashing (LSH)** to detect near-duplicates efficiently.

**2. Role-based classification**

A researcher, an AI engineer, and a founder have different information needs. I experimented with traditional ML, then tried JEV and the open-source alternative AlexaWortega/openjev.

After initial experiments and fine-tuning on sample articles I got some results but they were not that accurate,   
I eventually got access to JEV(waitlisted) and organised classification into four categories:

* AI / ML Researcher
* Software / AI Engineer
* Startups & Innovations
* Noise

The results have been encouraging.

**What's next?**

I want to build separate signals for relevance, potential impact, source reliability, and factual accuracy before ranking stories and delivering a personalized daily email.

My current pipeline is evolving from simple aggregation toward **a personalized intelligence layer for staying informed without drowning in information.**

I'm curious about three things:

* How would you combine MinHash/LSH with embeddings for better deduplication?
* How would you evaluate a role-based classifier when categories overlap?
* How would you rank stories by impact while keeping inference costs low?

document fetched 2026-10-03 00:18 · canonical https://reddit.com/r/AI_Agents/comments/1wvtzt9/how_do_you_keep_up_with_ai_when_new_things_happen