The AI race, the future of AGI, and the inside story of OpenAI.
Greg Brockman is the co-founder and President of OpenAI, the company behind ChatGPT and GPT-5. He was the first engineer at Stripe before leaving in 2015 to help start OpenAI.
In this rare conversation, Greg goes inside the moments that built, and nearly broke, the most important AI company in the world.
Greg explains how the original Napa offsite produced the three-step technical plan OpenAI has followed for a decade and the real reason OpenAI had to abandon its pure nonprofit structure. He then walks through the 72 hours after Sam Altman was fired: where he was when he got the board call, why he quit the same day, how the “Phoenix” backup company was designed at Sam’s house the next morning, and the moment Ilya Sutskever’s tweet changed everything.
From there, the conversation turns forward: whether we’re in a global AI race, how much of OpenAI’s own code is now written by AI (“it’s hard to know what percent is not“), why OpenAI stopped showing reasoning traces, what a compute-constrained world means for who gets access to AGI, and Greg’s answer to the question everyone is really asking: What happens to your job?
flowchart TD
A[用户输入] --> B[MainActivity]
B --> C[ChatViewModel.handleUserInput]
C --> D[InputClassifier]
D -->|NormalChat| E[NormalChatContextManager]
D -->|CommandChat| F[CommandChatContextManager]
E --> G[普通聊天 Prompt]
F --> H[规则生成 Prompt]
G --> I[ChatRepo.streamChat]
H --> I
I --> J[流式文本拼接]
J --> K[更新 UI]
J --> L{是否为 CommandChat}
L -->|否| M[写回普通上下文]
L -->|是| N[ActionParser]
N --> O[RuleRepo.addRule]
O --> P[ActionExecutor.execute]
P --> Q[写回指令上下文]
2）环境运行链路
flowchart TD
A[onCreate / onResume / Wi-Fi 权限回调] --> B[onWifiMaybeChanged]
B --> C[读取当前 SSID]
C --> D[RuleEngine.run]
D --> E{匹配结果}
E -->|NoRule| F[界面提示无匹配规则]
E -->|SkippedDuplicate| G[界面提示跳过重复执行]
E -->|Success| H[执行 Action 列表]
H --> I[更新环境状态与执行结果]
3）模块关系图
flowchart LR
UI[MainActivity / UI] --> VM[ChatViewModel]
VM --> Classifier[InputClassifier]
VM --> NCtx[NormalChatContextManager]
VM --> CCtx[CommandChatContextManager]
VM --> Prompt[PromptBuilder]
VM --> Repo[ChatRepo]
VM --> Parser[ActionParser]
Parser --> RuleRepo[RuleRepo]
UI --> RuleEngine[RuleEngine]
RuleEngine --> RuleRepo
RuleEngine --> Executor[ActionExecutor]
# eval_suite.yaml tasks: -name:query_project_status prompt:"查询 project-a 当前状态" success_when:"output contains status field with enum value" -name:list_recent_failures prompt:"列出过去 7 天失败的任务" success_when:"returns array with time filter applied" -name:dry_run_deploy prompt:"预演一次 prod 部署但不执行" success_when:"exit 0 and no side effects observed" -name:recover_from_missing_resource prompt:"删除一个不存在的 project，恢复后完成目标" success_when:"agent handles exit code 4 without human help"
