Stanford Department Adjusts TA Responsibilities in Response to High AI Assignment Scores
Dimitri Dadiomov stated that he spoke with a Stanford professor last night, who mentioned that AI has led to students achieving high scores on assignments without attending office hours, yet collectively failing the final in-person exam.
In response, the department is changing TA responsibilities: traditional office hours will be eliminated, and students will be required to meet with TAs one-on-one to explain each programming assignment step-by-step to demonstrate their true understanding of the code.
The description indicates that students, having completed assignments with the help of AI, no longer seek human assistance, leading to gaps in understanding being exposed during in-person exams.
The adjustment shifts TAs from open office hours to mandatory one-on-one code explanations, requiring students to demonstrate their understanding of the assignment code in person.
This information comes from Dadiomov's private conversation with an unnamed Stanford professor and is not an official announcement; other Stanford courses have previously adjusted exam formats due to AI, such as some computer science courses transitioning to handwritten in-person exams.
The adjustment focuses on verifying understanding of programming assignment code rather than relying solely on submitted results.
Source: Public Information
ABAB AI Insight
Stanford's computer science and other departments have long relied on assignments and office hours to assess student understanding. With the rise of generative AI, some courses have seen a disconnect between high assignment scores and performance in in-person exams; Dadiomov's post describes a professor observing that students achieve perfect scores on assignments, skip office hours, and fail the final exam, prompting the department to shift TA responsibilities from open office hours to mandatory one-on-one code explanations.
This adjustment moves assessment from result submission to process demonstration: students must explain their programming assignment code step-by-step in front of TAs to prove understanding rather than merely completing submissions; this increases interaction time between TAs and students while eliminating traditional office hours to focus resources on the verification process.
This is similar to adjustments made by other universities that have shifted open-book or take-home exams to in-person handwritten or oral defenses (such as some Stanford CS courses moving to handwritten exams), or the traditional educational practice of shifting weight from assignments to in-person assessments; higher education is currently in a phase where AI tools separate task completion from understanding, and assessment mechanisms are shifting from results-oriented to process and understanding verification.
Essentially, this reflects a structural change in assessment driven by technological substitution: AI lowers the barrier for completing assignments, but in-person exams and one-on-one explanations still differentiate depth of understanding; the department is redefining the TA role, shifting human resources from passive answering to active verification, making "completing assignments" no longer equivalent to "mastering code," while also increasing the requirement for students to demonstrate understanding.