aeon-paper-pick · diff
git:20260512.574b491 to git:20260512.a4ba040
28 added, 55 removed. Audit A to A.
---
name: aeon-paper-pick
description: |
- Surface the one AI / ML paper to read today from Hugging Face Papers, with a one-paragraph
- summary, the key claim, and why it's worth your hour. Filters out incremental work and
- benchmark-chasing in favor of papers with a falsifiable thesis or a method shift. Use as a
- daily morning brief input for AI-savvy operators.
+ Surface the one AI / ML paper to read today from Hugging Face Papers, with the central claim,
+ why it's worth an hour, where it might be wrong, and a time-budgeted read order. Filters out
+ pure benchmark-chasing and incremental scaling reports. Use as a daily morning brief input
+ for AI-savvy operators.
Triggers: "one paper to read today", "best AI paper today", "what's the must-read paper",
- "pick today's paper", "HF Papers top pick".
+ "HF Papers top pick".
---
# aeon-paper-pick
- One paper per day, picked from the HF Papers feed. Not a digest — one pick, with a short brief that lets the reader decide whether to invest the next hour reading the full PDF.
+ One paper per day. Not a digest — one pick, with a short brief that lets the reader decide whether to invest the next hour.
## How the pick works
- 1. **Ingest** — fetch the last 24h of Hugging Face Papers (`https://huggingface.co/papers`).
- 2. **Filter** — drop papers in three categories:
- - Pure benchmark-chasing (new SOTA on existing leaderboard, no method change).
- - Incremental scaling reports without a method delta.
- - Position papers without empirical claims.
- 3. **Score** the survivors on:
- - **Novelty** of the method or framing.
- - **Falsifiability** of the central claim (is there an experiment that would refute it?).
- - **Reproducibility signal** (code release? checkpoints? data?).
- - **Cross-discipline applicability** (does it port outside the lab that produced it?).
- 4. **Pick** the highest-scoring survivor. If two tie, the one with the sharper falsifiable claim wins.
+ 1. Fetch last 24h of HF Papers (`https://huggingface.co/papers`).
+ 2. Drop pure benchmark-chasing (new SOTA on existing leaderboard, no method change), incremental scaling reports, and position papers without empirical claims.
+ 3. Score survivors on novelty of method/framing, falsifiability of central claim, reproducibility signal (code + weights + data), and cross-discipline applicability.
+ 4. Pick the highest scoring. Tie-break: sharper falsifiable claim.
- ## What the brief contains
+ ## Brief format
```
- *Paper Pick — 2026-05-12*
-
[Paper title]
Authors: A, B, C (Lab)
arXiv: 2505.xxxxx
- HF Papers: huggingface.co/papers/2505.xxxxx
- The central claim (1 sentence, plain English):
- The authors argue that X causes Y when Z, demonstrated on dataset W.
+ Central claim: [one sentence, plain English]
Why it's worth an hour:
- - It's a method shift, not a benchmark bump — they replace [common technique] with [new technique].
- - The claim is falsifiable: ablation in §4 isolates X vs not-X cleanly.
- - Code + weights released — reproducible at home with a single GPU.
+ - [method shift, not benchmark bump — they replace X with Y]
+ - [claim is falsifiable: ablation in §4 isolates X vs not-X cleanly]
+ - [code + weights released, reproducible at home]
Where it might be wrong:
- - The training distribution is narrow (only dataset W). Open question whether the effect transfers.
- - The comparison baseline is older than expected — would a 2025-tier baseline still trail?
+ - [training distribution is narrow — open question whether the effect transfers]
+ - [comparison baseline is older than expected]
Read order:
- 1. §3 (method) — 8 minutes
- 2. §4 (ablations) — 12 minutes
- 3. §6 (limitations) — 5 minutes
+ 1. §3 (method) — 8 min
+ 2. §4 (ablations) — 12 min
+ 3. §6 (limitations) — 5 min
Optional: §5 (extended experiments)
```
- ## When this surfaces nothing
+ ## Empty days
- If the day's filter passes 0 papers (rare but happens on slow days), the output is:
+ If 0 papers survive the filter:
```
- *Paper Pick — 2026-05-12*
-
- No surviving picks today. 14 papers were filtered out:
- - 6 benchmark-chasing
- - 5 incremental scaling
- - 3 position papers
-
+ No surviving picks today. 14 papers filtered out: 6 benchmark-chasing, 5 incremental scaling, 3 position papers.
Worth scrolling yourself: huggingface.co/papers
```
Honesty over manufactured picks.
- ## Guidelines
-
- - One pick per day, not three. Three picks = digest, and digests are wallpaper.
- - "Why it might be wrong" section is mandatory. No paper is bulletproof.
- - Read order is required and time-budgeted. Save the reader from skimming the whole PDF.
- - Cite the actual sections (§3, §4) so the reader can verify the brief.
-
- ## Pairs with
-
- - `aeon-huggingface-trending` for breadth (models, datasets, spaces).
- - `aeon-deep-research` when the paper sparks a research thread.
- - `aeon-last30` to check whether the paper's claim is being discussed.
-
- ## Required keys
+ ## Rules
- None — Hugging Face Papers is public.
+ - One pick, not three. Three picks is a digest, and digests become wallpaper.
+ - "Where it might be wrong" is mandatory.
+ - Read order time-budgeted. Save the reader from skimming the whole PDF.
+ - Cite actual sections so the reader can verify the brief.