HiddenState – How I keep up with 500+ ML papers a day

Name: HiddenState – How I keep up with 500+ ML papers a day
Availability: InStock
Author: CosmoSantoni

by CosmoSantoni·Feb 18, 2026·1 point·1 comment

Visit Project View on HN

AI Analysis

●●SolidBig BrainNiche Gem

The Take

Clustering by the specific technical constraint being attacked — not by topic — and scoring each signal on convergence, implementation evidence, engagement and significance is a neat, high-signal trick for surfacing research trends. It smartly dedupes org noise and ingests many sources, though using Claude as a clustering black box means the scoring pipeline could use clearer auditability or export hooks for skeptical researchers.

Post Description

HiddenState monitors arxiv, Reddit, GitHub, HN, Bluesky, HuggingFace, OpenReview, PapersWithCode, and a handful of research blogs. Every few hours it pulls new items, throws most of them away (+95%), and clusters what survives by the specific technical constraint being attacked. Not by topic, not by domain.

Example from this week; 7 independent VLA papers dropped within 24 hours from 9 different orgs. Xiaomi, GigaBrain, RISE, all attacking sim-to-real transfer for robotic manipulation. None coordinating. That kind of convergence is hard to spot unless you're reading everything.

Each mechanism gets a 0-100 score across convergence, implementation evidence, engagement, and significance. Orgs are deduplicated so a single lab posting on five platforms doesn't inflate the signal.

Python, SQLite, Claude for clustering, Cloudflare Pages. Free, no tracking. Looking for any and all feedback and thoughts! Cheers!