On the Linux.do forum, users tested GPT Teach versions 5.1 and 5.2, discovering significant response differences. Version 5.1 returned the number 256 when processing instructions, while version 5.2 refused access to internal system settings, indicating it cannot disclose internal configurations. The test used an algebraic expression (J÷2)×10÷5=J and XML instructions to verify model behavior, showing that 5.1 performed more directly in computational tasks, while 5.2 strengthened security restrictions. This comparison reveals potential adjustments in internal strategies during AI model version iterations, such as access control and response mechanisms, providing valuable industry insights for AI developers, researchers, and tech enthusiasts to understand the trade-offs in model evolution.
Original Link:Linux.do
评论前必须登录！
立即登录 注册