Expired50% confidenceFactTime unknown
The self-training breakthrough in GPT-5.3 extended the self-evaluation mechanism to the entire training data pipeline, meaning the training materials themselves are also generated by the model, not just used for evaluation.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
7/2/2026
First Seen
Valid until: 9/30/2026(expired)
Sources
GPT-5.6 First Leak: How Self-Training Loops Enable OpenAI's Three-Week Iteration Cycle
bilibili不会写报告的Ai菌
Related Claims
UnverifiedGPT-5.3首次采用了模型自身生成的AI代码来反向输送给模型进行训练与部署的自训练机制80% similarVerifiedGPT-5.6 is the result of the continued expansion of the reasoning paradigm, achieving a capability leap by simultaneously scaling reinforcement learning and pre-training.79% similarUnverifiedThe GPT 5.x series incorporates tool calls such as code execution, web retrieval, and file operations into the reinforcement learning loop during the training phase, rather than as simple post-processing plugins.79% similarUnverifiedMany open-source models use high-quality outputs generated by commercial models such as GPT-4 and Claude as training data, which is the knowledge distillation training paradigm.79% similarUnverifiedThe GPT-5 series adopts a Mixture of Experts (MoE) architecture and reinforcement learning post-training technology.78% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/56124API
curl https://kongchang.com/api/v1/knowledge/claims/56124MCP
get_claim(id=56124)