Yan Junjie

Yan Junjie, also referred to as I.O. in the source, is the founder and CEO of MiniMax. In 对话 MiniMax 闫俊杰：M3、10X 计划、10T 模型、和智能的终局, he explains MiniMax’s model progression, argues that AI’s core value is productivity, and frames model progress, agents, domain experts, scaling, and interpretability as connected parts of the same intelligence problem.

Key Views

• AI coding should expand production capacity and make more people productive, not be reduced to a simple headcount story.
• M1, M2/M2.7, and MiniMax M3 show how model capability, token usage, coding feedback, and product expectations interact.
• Model Harness Co-Evolution matters because agents and harnesses are not merely downstream from models; real workflows also shape model requirements.
• Frontier Model Scaling is not blocked by one magic obstacle, but by time, compute, architecture, training efficiency, and data limits.
• Domain Expert Alignment is needed when AI moves into safety, finance, law, and other areas where domain judgment cannot come only from researchers or engineers.
• AI Interpretability By AI is a long-term safety question: stronger AI may be needed to help humans understand AI itself.

Connections

• MiniMax and MiniMax M3 — company and model context.
• AI Coding Verification — engineering bottleneck surfaced through AI coding adoption.
• AI Inference Cost Structure — token usage and model cost context.
• Anthropic — referenced as an example of bringing non-engineering experts into frontier AI work.