Large Language Models (LLMs) are increasingly employed as AI tutors due to their scalability and potential for personalized instruction. However, off-the-shelf LLMs often underperform in educational settings: they frequently reveal answers too readily, fail to adapt their responses to student uncertainty, and remain vulnerable to emotionally manipulative prompts. To address these challenges, we introduce CoDAE, a framework that adapts LLMs for educational use through Chain-of-Thought (CoT) data augmentation. We collect real-world dialogues between students and a ChatGPT-based tutor and enrich them using CoT prompting to promote step-by-step reasoning and pedagogically aligned guidance. Furthermore, we design targeted dialogue cases to explicitly mitigate three key limitations: over-compliance, low response adaptivity, and threat vulnerability. We fine-tune four open-source LLMs on different variants of the augmented datasets and evaluate them in simulated educational scenarios using both automatic metrics and LLM-as-a-judge assessments. Our results show that models fine-tuned with CoDAE deliver more pedagogically appropriate guidance, better support reasoning processes, and effectively resist premature answer disclosure.

A study of 100 people from a variety of ages and technical aptitudes conducted at business school Bentley University in Massachusetts found that just as with other humans, people form first impressions of chatbots that stick. The study was conducted as part of the school's Human Factors and Information Design program, by students working under Bentley adjunct lecturer Meena Kothandaraman, founder of the twig fish research practice, and in partnership with NeuraFlash, a Boston-based firm that uses AI for Salesforce consulting. The takeaway for botbuilders is obvious: However sophisticated your bot, or whether it hangs out with Maroon 5, its first greeting and responses to new contacts will determine whether they find it an automated annoyance or a new best friend.