Apodex

Apodex is the AI company discussed in E242｜最快半年AI跑通自进化？与陈天桥首席科学家聊聊硅谷模型必争之地. The source presents it as a Chen Tianqiao-backed and strategically directed company focused on Heavy Duty Solver work rather than consumer chat, image generation, or video generation. Its target is a Discovery Model: a system that can propose hard scientific hypotheses, search and reason about them, write code or run simulations, and verify whether the hypotheses hold.

The episode describes Du Shaolei as Apodex’s chief scientist for training and inference, and Li Beibin as chief scientist for code and self-evolution. Their shared technical path connects Deep Research, agent-team problem solving, AI Verification, Recursive Self-Improvement, and Research Taste. The company is also described as using a model post-trained from Qwen for Deep Research-style capabilities.

Source Position

• Apodex wants to solve hard problems directly rather than only sell tokens or build another chat assistant.
• Its near-term path is a post-training and harness loop: diagnose model weaknesses, construct tasks and training data, train, verify, and adjust the harness.
• Its verification strategy includes code tests, formal math verification where possible, and Multi-Agent Collaboration when no single ground truth exists.
• Its first high-value domains in the source include biology, medicine, drug discovery, old-drug repurposing, and diagnosis.
• Chen Tianqiao is described as shaping both strategic direction and model behavior through a constitution-like set of priorities.

Connections

• Chen Tianqiao, Du Shaolei, and Li Beibin — founder/strategic leader and chief scientists in the source.
• Recursive Self-Improvement, Discovery Model, Deep Research, and AI Verification — main technical themes.
• Research Taste, Problem Definition In Research, and AI For Science — scientific-discovery constraints.
• Qwen — base model family named for the post-trained model in the episode.