Shreya Shankar's group publishes 5 rules for readable bulk AI text
sh_reya · x · 2026-11-01
Shreya Shankar's group at UC Berkeley published "Five Principles for Generating Readable AI Text in Bulk," addressing why agent session titles, summaries, and dashboards are so hard to skim. The rules, grounded in decades of reading research:
- Start every item with its subject or verb (readers take the first noun as the subject)
- Put the words that tell items apart first (skimmers only see the first two words; Amazon writes "Ordered:" before the product name)
- Use one structure and one word per concept (Notion AI's summaries opened three different ways)
- Vary items only when the difference means something (a version label on just one card implies importance)
- Pick verbs that show good or bad news ("Users love" vs "Users complain")
The authors suggest handing the checklist to your coding agent so it generates readable titles and summaries.
More from coding & agent
- Scholar releases skill fixing ~27 AI rhetorical tics in Codex-written surveys — tak3sh8 · 2026-11-01
- AI Memory as Lock-In: One User Keeps Memories in Own Files via Hermes Agent — alexcovo_eth · 2026-11-01
- ISUCON 2026 finalist says the tuning contest became a whole different game with AI — tkasasagi · 2026-11-01
- 64GB Macs are no longer a dead zone for agentic coding: 27B model at 256k context, measured — M5_Maxxx · 2026-11-01
- Together AI: route each task to the right model size instead of defaulting to the biggest — togethercompute · 2026-11-01
- How Boardy made its CI 20x faster, cutting PR waits from 40 minutes to 2 — HankYeomans · 2026-11-01