Should I read this HN thread? Claude TL;DR

Pull every comment on an HN story via Algolia, flatten the tree, hand to claude for a dominant-view + dissent summary. Saves an hour.

$ ID=42583107   # any HN story id
curl -s "https://hn.algolia.com/api/v1/items/$ID" \
  | jq -r '.. | .text? // empty' \
  | claude -p "These are all comments on an HN story. In 5 bullets: dominant view, strongest dissent, any concrete numbers cited, who got it right historically (if anyone), one open question."

1. [01] curlcurl -s "https://hn.algolia.com/api/v1/items/$ID"
   Algolia's `items` endpoint returns the FULL comment tree of a story as nested JSON — children inside children inside children.
2. [02] jqjq -r '.. | .text? // empty'
   The recursive-descent trick: `..` walks every node, `.text?` returns the field if present, `// empty` skips nulls. Result: one comment text per line, tree flattened.
3. [03] claudeclaude -p "In 5 bullets: …"
   All comments concatenated arrive on stdin. Claude doesn't need the tree structure — just the prose.

- **Dominant view**: most commenters agree the new tokenizer is the real story, not the model size.
- **Strongest dissent**: @user42 argues the benchmark is gameable and points to a 2024 leak; gets 38 upvotes.
- **Numbers cited**: 3.2x faster inference, 41% lower latency P95, $0.12/1M tokens for the open-source variant.
- **Got it right**: @user17 predicted this approach 4 months ago in a now-linked comment.
- **Open question**: nobody explains how it handles non-Latin scripts.

• Big threads (>1000 comments) may exceed claude's context — slice with `head -c 200000` before piping.
• The Algolia items endpoint is the ONE source that gives you the full tree in a single call — the firebase API requires recursive fetches.