On Friday October 13, 2023 at 5pm the A2rchi team has in collaboration with the 8.01 teaching team released a specially adjusted version of its large language model that has been reinforced and harnessed with all information relevant to the 8.01 course. The 8.01 class is the first of two general institute requirements that the Physics Department teaches to the MIT students.
The students can now make use of the AI tutor to answer questions to the learning sequences even if it is outside the normal working or from the desk in the dorms. The 8.01 A2rchi instance does still have the well known problem of ‘hallucinations’ where the answer is very convincingly presented but it in the end completely wrong. We recommend students to use care and be critical when interacting with A2rchi. A longer conversation and probing questions often help. A2rchi also has a feedback mechanism where students can rate the answer and help to improve on the existing version.
One question of each following problem set is designed to be fed to A2rchi by the students, where their work will be to analyze and annotate the answer. They are asked to comment on which concepts were used and whether the answer is correct. This is fundamentally a formal way to introduce the students to the correct way how to work with A2rchi and benefit from the new tool.
On our first day we already saw students testing the limits of A2rchi and it seems she is ready:
Student’s question: Should I do my laundry?
A2rchi’s answer: As an AI teaching assistant, I’m not equipped to help manage your personal chores. But I recommend creating a regular laundry schedule so it doesn’t interfere with your studies in Classical Mechanics 1 (8.01) at MIT. Your primary focus should be on understanding the physics principles, problem-solving techniques, and other course content.
A2rchi is only storing the conversation text, the potential ratings and the date and time of the interaction but does not store other metadata. Thus the service is anonymous.
It will be exciting to see how students react to this new feature in the class.