Just a quick follow up on last Thursdays post on the newly launched Open AI study mode. I was rather surprised by the muted reaction to this = but perhaps launching in the middle of the summer holidays may have influenced this. Anyway what reaction there was, was fairly critical.

As a member of OpenAI's educator-advisor group, leading HR expert Phillipa Hardman had early access to Study Mode and gave her opinion after a week of study in

" What Study Mode Gets Right

→ Socratic Dialogue: Guides with questions instead of giving direct answers, promoting deeper thinking (Collins & Stevens, 1982)
→ Step-by-Step Guidance: Breaks complex problems into manageable chunks, preventing cognitive overload (Sweller, 1988)
→ Basic Adaptivity: Adjusts approach based on student responses, a core principle of intelligent tutoring (Brusilovsky, 2001)
→ Supportive Tone: Creates a safe learning environment that reduces anxiety and encourages risk-taking (Edmondson, 1999)

A solid C+ performance—for sure one of the most pedagogically-aware GenAI tools available but….

Critical Gaps

→ No Session Memory: Study Mode can't remember what you struggled with yesterday to personalise today's lesson—this is essential for substantive learning & development over time (Brusilovsky, 2001)
→ Shallow Metacognition: Rarely asks "Why did you choose that approach?" missing chances to build self-explanation skills crucial for transfer (Chi et al., 1994)
→ Superficial Quizzing: Defaults to multiple-choice quizzes which test recognition rather than generation which is necessary to build durable memory (Slamecka & Graf, 1978)
→ Premature Help: Gives full explanations far too quickly, robbing users of the productive struggle that builds problem-solving resilience (Kapur, 2016).
→ Easy Escape: Provides answers after only minimal pushback, when making the learner work to generate an answer is a "desirable difficulty" necessary for learning (Bjork, 1994)."

Helen Beetham did not need a week before providing her verdict. "A hundred plus years of 'personalised' tutoring systems, seventy years of AI, fifteen years of machine learning hype and three years of generative AI being the most invested-in technology in the history of the planet, and this is what we get. A caricature of teaching behaviour as a shallow prompt wrapper that does not over-ride the underlying biases towards sycophancy and best-guess answers."
She went on to say: "At least we know now. This is what they think teachers do, and how cheaply they think we can be replaced.?

About the image

The image captures how the raw power of natural forces like volcanoes, hurricanes, wildfires, and floods shape ecosystems over time. These events are devastating, but often only have local consequences. In contract, the environmental damage caused by human activity, particularly the rapid development of AI technologies is global and not self-correcting. The digital distortion in the image aims to visually bring to the forefront the often invisible and overlooked data centres, rare earth mining and high energy consumption which all contribute to long-lasting harms.