DINQ Session AI salon kicks off: Why OPD is becoming infrastructure for the LLM era | BlueRun Ventures

DINQ Session, launched by BlueRun Ventures portfolio company DINQ, is now officially open — with BlueRun Ventures joining as a co-supporter. DINQ will host monthly in-person gatherings for AI researchers, bringing together academics, developers, and founders to exchange ideas on cutting-edge topics including AI agents and model algorithms.

The inaugural session will take place on September 20 in Beijing's Wudaokou neighborhood, focusing on On-Policy Distillation (OPD). Researchers and practitioners from Tsinghua University, Stanford University, Renmin University of China, and The Hong Kong Polytechnic University will join the discussion, covering everything from model training and Multi-Teacher approaches to cross-vocabulary distillation, agents, and scalable intelligence.

As large models continue to advance, how to make them learn, iterate, and scale more efficiently is becoming an increasingly pressing question. We welcome you to join us on-site to discuss OPD and the evolving landscape of large model training. Registration details below.

Speakers & Topics

1. Bingxiang He

PhD Candidate, Tsinghua University NLP Lab

Topic: Rethinking On-Policy Distillation of Large Language Models: Getting it started, and what happens after

2. Xinyu Tang

PhD Candidate, Gaoling School of Artificial Intelligence, Renmin University of China / Lead, Ring Agent

Topic: From Inside the Model to Beyond: A Two-Layer Architecture for Scalable Intelligence; presenting papers Ring Zero and Harness Survey

3. Anhao Zhao

PhD Candidate, The Hong Kong Polytechnic University NLP Group

Topic: Multi-Teacher OPD & Cross-Vocabulary OPD

4. Jiaqi Chen

Visiting Scholar, Stanford University

From Presentations to Face-to-Face Discussion

Beyond the speaker sessions, this gathering will include roundtable discussions and open networking.

From OPD to large model training, from research methodologies to real-world applications, we aim to bring together diverse research directions and practical experiences to discuss where the technology is heading and what questions remain worth exploring.

If you're working on related research, projects, or implementations, we encourage you to bring your questions to the room.

Event Details & Registration

Organized by DINQ, with support from Elsewhere, Founder Park, BlueRun Ventures, Yuan Alumni Association, and Volcano Engine

「📅」 Date: September 20 (Friday)

「⌚️」 Time: 13:30–17:00

「📍」 Location: Beijing · Wudaokou

(Exact address provided upon registration)

「👥」 Capacity: Limited to 80 attendees

「🎫」 Cost: Free admission, with refreshments provided

「🔗」 Registration: https://luma.com/mbcy2f9i

Scan the QR codes below to join the event WeChat group

and receive updates on future sessions.

[Luma Registration QR] [Group Chat QR]

If you're following developments in large model training,

model distillation, agents, or scalable intelligence,

we welcome you to join researchers and practitioners on-site

to discuss the technical shifts underway.

September 20. See you in Wudaokou, Beijing.

About DINQ Session

A monthly AI salon series,

focused on the frontier of emerging technology.

Bringing together academics, developers, and founders

for deep discussions on AI agents, model algorithms,

and commercialization —

bridging research and industry,

sparking ideas, and connecting like-minded builders.

VLA and World Models: Did This 4-Hour Technical Salon Reach Any Conclusions?

BlueRun Ventures AI Annual Review: Survival, Evolution, and Narrative for China's AI Founders | Booming Talk

2026 AI Inflection Points: How Will OpenClaw Shape AI Entrepreneurship? What's Changing in Models and Applications? | BlueRun Ventures AI Annual Review