AI Tutor
An AI system that provides assistance to students in educational settings. Its central design question concerns the help strategy: whether to give answers directly or to guide learners to attempt the problem first. Research indicates that different strategies have markedly different effects on learning transfer.
Timeline (last 90 days)
The core architectural rule of the AI tutor system is: the LLM never decides whether to help or how much to help—that is decided by ordinary code; the LLM is only responsible for writing out the wording of the hint after the decision has been made.
The first version of the attempt-judging logic "a result set different from the previous one counts as an attempt" had a bug: queries that failed to run had no result set, so for learners who could not get anything to execute no matter how they wrote it, the door never opened.
The study recommends shifting from a questioning mode to a direct error-correction mode for deep impasses to avoid wheel spinning, rather than adopting a one-size-fits-all guardrail design that prohibits giving answers.
步步高在AI大模型能力（如AI对话辅导、作文批改智能程度等）方面的口碑弱于科大讯飞等以AI技术见长的竞品，选购时若重视AI智能辅导深度需对比体验
All Facts (3)
The core architectural rule of the AI tutor system is: the LLM never decides whether to help or how much to help—that is decided by ordinary code; the LLM is only responsible for writing out the wording of the hint after the decision has been made.
50%UnverifiedThe first version of the attempt-judging logic "a result set different from the previous one counts as an attempt" had a bug: queries that failed to run had no result set, so for learners who could not get anything to execute no matter how they wrote it, the door never opened.
50%UnverifiedThe study recommends shifting from a questioning mode to a direct error-correction mode for deep impasses to avoid wheel spinning, rather than adopting a one-size-fits-all guardrail design that prohibits giving answers.
50%