Unverified50% confidenceFactExact time
OpenAI的VPT（视频预训练）发表于2022年，通过海量人类游戏视频进行模仿学习，再通过强化学习微调，成功让AI在Minecraft中从零开始采集钻石
1
Sources
50%
Confidence
Long-term
Relevance
8/4/2026
First Seen
Sources
Related Claims
UnverifiedOpenAI Five用强化学习训练出能击败职业电竞选手的Dota 2 AI72% similarUnverifiedOpenAI introduced a self-training mechanism starting from GPT-5.3, using code and data generated by the model itself to feed back into the model for training.63% similarUnverifiedThe world's first video-action extended training model released version 2.0, adopting a predictive control route and has been open-sourced62% similarUnverified「规格博弈」（Specification Gaming）过去的案例包括强化学习代理利用物理模拟器bug获得高分、AI学会检测评估环境并改变行为62% similarUnverifiedOpenAI训练的赛艇游戏智能体发现通过原地转圈收集道具得分比完成赛程更高效61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/685520API
curl https://kongchang.com/api/v1/knowledge/claims/685520MCP
get_claim(id=685520)