pi-vcc vs pi-blackhole: Sub-second Lossless Context Compaction or 3-LLM Recall?
mailto_devnull · reddit · 2026-09-14
The author discovered pi-vcc two days ago: model-free, sub-second context compaction with near-lossless recall, which works great. Then found pi-blackhole, which instead uses three small LLMs to build better recall. They question whether moving away from non-deterministic LLM summaries just to call models again is worth it, and ask for comparisons — currently leaning toward staying on pi-vcc.
More from coding & agent
- Épi self-evolving research framework nets 120 record results in days — my_cat_can_code · 2026-09-14
- Minecraft is now too big for RL: dev suggests training on older versions instead — gandamu_ml · 2026-09-14
- Open-source ComfyUI nodes bring relight video generation via MiniMax H3 — Emotional_Example_12 · 2026-09-14
- CUDA-accelerated Minecraft worldgen hits bit-accurate parity and 9-player splitscreen — gandamu_ml · 2026-09-14
- Ran an AI Agent for 2 Days and 300% of Quota to Fix My App — Sign-in Still Broken — jarrodwatts · 2026-09-14
- Hugo Bowne launches Master Agentic Data Science workshop with PyMC's Thomas Wiecki — hugobowne · 2026-09-14