A dot-and-bang tag system to control LLM answer depth and length
viktor_zinchenko · reddit · 2026-09-18
Tired of LLM essays when you just want a quick answer, the author built a minimal zero-shot tag system using only . (length) and ! (depth):
- Depth: ! = newcomer level (analogies, simple explanations); !! = balanced practitioner; !!! = expert (precise terminology, deep dive)
- Length: . = answer only (1-3 sentences or one code block); .. = answer plus minimal reasoning; ... = detailed (context, trade-offs, edge cases)
Prefix prompts like ..!!! and you're set. Rules: no tags or missing axis = balanced; tags only change form, not honesty/language behavior; explicit natural-language instructions always override tags. Designed to be fast on both mobile and desktop keyboards; drop it into custom instructions.
More from Apps
- Obsidian Starter Kit v4 ships with MCP server, osk-cli and ~375 specialized AI skills — dSebastien · 2026-09-18
- New 'jev board' keyboard experiment explores ambient-intelligence interfaces powered by fast models — ojasvi_yadav · 2026-09-18
- Indie teleprompter app rolls out voice-following improvements, iPhone Duo on roadmap — yihui_indie · 2026-09-18
- OutlierKit launches API and MCP to turn AI agents into YouTube strategists — ayushtweetshere · 2026-09-18
- Open-source Chrome extension skips YouTube sponsor reads in real time for ~$0.005 per video — yangyi · 2026-09-18
- Scale AI founder Alexandr Wang touts docking Muse to change your workflow — alexandr_wang · 2026-09-18