Create

Sign in to ReadmeX

or

How we handle your data: Privacy Policy

ReadmeX

Your AI reading companion

What would you like to catch up on?

Pick a briefing, or ask your own question.

Your community and people briefings stay personal to you.

Story

BIGAI-incubated Tongyan Lingzhi bets on recursive self-improving personal AI

AI summary

Quantum位 reports that Tongyan Lingzhi, a company drawing on the Beijing Institute for General Artificial Intelligence (BIGAI) ecosystem, is developing a Personal AI built around parameterized learning and continual evolution: lightweight Personal LoRA adapters sit on top of a shared base model to carry the preferences of individuals, projects and enterprises, while an RSI loop uses an Evaluator to filter feedback, re-runs updated models against the previous version on the same task batch, and rolls back regressions. The company says its core RSI technical route and product architecture are defined, with modules moving into engineering, while a fully automated closed loop is still under development. In staged tests using Qwen2.5-72B-Instruct and eight NVIDIA H800 GPUs, Paged and conventional Flat LoRA modes were both held to a 5.79GB memory budget; Paged showed some latency and tail-latency advantage at 1%–30% LoRA request share, but frequent parameter swapping still degraded performance at 50%. These architecture, mechanism and test results are the company's own claims and have yet to be validated in production.

In-depth insight

The report frames parameter-level personalization plus a self-improvement loop as a way for organizations to turn long-term business experience into their own model capability rather than depend on a single base model, though the technology and tests still await industry validation.

通研院孵化的团队,也盯上了RSI – 量子位

< img id="wx_img" src="https://www.qbitai.com/wp-content/uploads/imgs/qbitai-logo-1.png" width="400" height="400">

通研院孵化的团队,也盯上了RSI

思邈 2026-10-19 22:31:12 来源:量子位

当AI越来越懂你,如何让智能真正属于你?

#Tongyan Lingzhi#Beijing Institute for General AI (BIGAI)#Qwen2.5-72B-Instruct#NVIDIA H800#OpenAI#Meta

Original · 16 min readRead the original
Group chat

No comments yet. Start the conversation.