BIGAI-backed startup unveils Personal AI stack: Personal LoRA, RSI loop and PagedLoRA serving
机器之心 · wechat · 2026-10-19
Tongyan Lingzhi, a startup backed by BIGAI (Beijing Institute for General Artificial Intelligence), detailed its Personal AI approach: instead of retraining base models, it carries per-user/project expertise in lightweight Personal LoRA adapters on a shared base model.
- Beyond prompt/RAG/memory: bolt-on personalization lets models "look up" but never "learn" implicit judgment; the startup moves personalization into parameters.
- RSI loop: an Evaluator learned from real Accept/Reject/review feedback curates training data, incrementally updates Personal LoRA, re-evaluates, ships or rolls back — and improves its own update strategy each cycle.
- TIP (train-inference parallelism): folds incremental training into online GPU scheduling to change the cost structure of continual learning.
- PagedLoRA: OS-style paging for adapters; on Qwen2.5-72B with 8×H800 under an identical 5.79GB LoRA budget, P50 latency beats flat multi-LoRA serving at 1–30% LoRA request ratios (2.2× faster metadata construction), with degradation identified above 50% load and async page-in planned.
More from Companies & People
- KU Leuven and VIB open two Group Leader professorships in AI for Biology, start 2027 — jpillowtime · 2026-10-23
- Google's Gemini pretraining team is hiring research engineers in London and MTV — zacharynado · 2026-10-23
- After data and compute, the next picks-and-shovels are talent and physical infrastructure — philhchen · 2026-10-23
- Critics slam Anthropic for rarely defending its work against criticism on Twitter — DKokotajlo · 2026-10-23
- Amazon blocks AI shopping agents while Walmart opens its site to them with spend controls — venturetwins · 2026-10-23
- Commenters speculate Skydance could wire Grok into its entire Hollywood production pipeline — MickeySteamboat · 2026-10-23