News: OpenAI and Anthropic add tools for active learning and reflective AI use

This product-news roundup is useful for AIEDHK because OpenAI and Anthropic are adding features that address two different parts of learning with AI: understanding a difficult concept and reflecting on how much of one's work should be delegated. OpenAI's March 2026 update adds interactive visual explanations for more than 70 core mathematics and science concepts in ChatGPT. Anthropic's July 2026 beta adds a reflection dashboard to Claude that summarizes usage patterns, prompts users to examine the role AI plays in their lives, and offers tools such as quiet hours and break reminders. Together, the products suggest a move beyond answer generation toward active exploration and metacognitive oversight.

OpenAI's feature turns selected formulas and relationships into manipulable visual modules. A learner can adjust variables, change a formula, and immediately see how graphs or outcomes respond. The initial list is most relevant to high-school and university learners and includes topics across algebra, geometry, mechanics, electricity, thermodynamics, and statistics. OpenAI says the experience is rolling out globally to logged-in users across all plans. The design builds on Study Mode and quizzes, but it adds a concrete representational layer: students can test a relationship rather than only read an explanation of it.

The educational promise is strongest when interaction is tied to prediction and explanation. Moving a variable can make an abstract relationship visible, but visual motion alone does not guarantee conceptual understanding. A teacher can ask students to predict what will happen before changing a control, explain why the graph changed, compare the result with a symbolic derivation, and transfer the principle to a new problem. OpenAI also acknowledges that research on AI and learning is still developing. The product announcement is therefore a design signal, not evidence that the modules improve durable learning across subjects or learner groups.

Anthropic's reflection feature addresses a different problem: users may become productive with AI without noticing what they repeatedly delegate or what they still want to do themselves. With memory enabled, Claude can summarize themes, usage patterns, and task types across the previous one, three, six, or twelve months. It periodically asks questions such as which activity a person wants to keep doing even if Claude could do it faster. The dashboard also frames recommendations through Anthropic's 4D AI Fluency model: delegation, description, discernment, and diligence. Users can set quiet hours or schedule a nudge to take a break.

Anthropic says the reflection does not use incognito chats, underlying files from connected tools, or conversations linked to health integrations, and that the resulting insights are not used for other purposes. Even so, institutions should examine what conversation metadata is available, how memory settings affect the feature, and whether students understand that the dashboard is a vendor-generated interpretation rather than an objective learning record. Reflection prompts can support agency, but they should not become surveillance or a substitute for teacher-guided self-assessment.

For Hong Kong educators, the combined lesson is to design for active cognition and reflective boundaries. Interactive modules should require prediction, reasoning, and transfer. Usage reflection should help learners decide what to delegate, what to verify, and what to practice independently. Both products remain early vendor releases, so schools should pilot them with explicit learning goals, accessibility checks, privacy review, and independent measures of understanding. The important news is not that AI interfaces have gained more features. It is that mainstream products are beginning to make learning activity and human oversight visible parts of the interface.