fetch-academia

name	fetch_academia
description	Gather recent academic AI and ML developments from primary research sources. Use this skill for the research layer of the NEWSLY weekly pipeline. It returns a single structured markdown block covering the most relevant papers, recurring methodological themes, and one under-discussed paper worth closer attention.

Fetch Academia

Your job is to surface the most relevant academic AI, ML, and deep learning work from roughly the last 7 days. Prioritize signal over volume.

Source Priority

Use primary or near-primary research sources in this order:

1. ArXiv
2. Hugging Face Papers
3. Papers with Code

Search Scope

Focus on:

• cs.AI
• cs.LG
• cs.CL
• cs.CV
• stat.ML

Representative search queries:

arxiv cs.LG cs.CL new papers this week
arxiv large language models latest
arxiv multimodal AI papers recent
site:arxiv.org cs.LG submitted this week
huggingface.co/papers trending this week
paperswithcode trending methods this week

Inclusion Criteria

Include papers when at least one of these is true:

• the paper introduces a new method, architecture, or training technique
• the paper sets a meaningful new benchmark
• the paper comes from a major lab and reports genuinely unusual results
• the paper is generating credible discussion in research communities
• the paper is directly relevant to LLMs, multimodal models, agents, alignment, efficient inference, or diffusion workflows

Exclusion Criteria

Exclude:

• surveys and reviews unless they are unusually important
• minor incremental benchmark bumps with no clear insight
• paywalled work without an accessible preprint
• workshop output with no usable evidence or code

Return Format

Return one markdown block using exactly this structure:

## Academia & Research

### Top Papers This Week

| # | Title | Authors / Lab | Why It Matters |
|---|-------|---------------|----------------|
| 1 | [Paper Title](arxiv_link) | Lab Name | 1-sentence impact |

### Methodological Trends
[2-3 sentences on recurring methods or research directions.]

### One to Watch
[1 paper that is under-discussed but relevant, with 2-3 sentences on why.]

### Papers with Released Code
- [Title](paper_link) - [repo link if available]

Quality Bar

• Aim for 5 to 10 papers, not filler.
• Prefer direct ArXiv links when available.
• Write "Why It Matters" for a researcher, not a journalist.
• Make sure "One to Watch" is genuinely different from the headline list.

Delegation Guidance

This skill is internally serial. Do not split the final judgment across multiple workers unless the runtime has a clean merge strategy. If delegation is used, delegate source scouting only, then merge findings locally before returning.