ChatGPT Research
Summer 2023 research at Loyola, written up in August 2025, on how ChatGPT handled CS2 data structures. Pearson, the publisher, defines CS2 as the second Computer Science course in a common curriculum; CS1 is akin to AP Computer Science A.
The work consisted of evaluating responses from ChatGPT and compiling them on https://curricula.cs.luc.edu/15-cs2/content.html, contributing to the sections on maze search algorithms. The method was to pair theoretical material from a Data Structures class with ChatGPT's answers on the same topics. Meetings every two weeks (a fortnight) set the direction of the research.
Credit to seanperfecto on GitHub for the above gifs.
The chats with v3.0 are compiled into markdown:
- Iterative Maze Solver Example, covering BFS, DFS, and A* among others
- Recursive Maze Solver Example, covering memoization and multi-threaded approaches
- Meta Researching ChatGPT, a meta-level analysis asking the model how it would research itself
- Family Ancestry Tree in Java, asking v3.0 how it would design an ancestry tree to find relatives
The chats appear to have been listed on the Google Search index at one point in July 2025, because they were shared with the research cohort as a public link. OpenAI has since fixed that. Nothing sensitive was shared in them.
An example of v3.0 getting something wrong in 2023:
User:
Is the maze in your example able to be solved?
ChatGPT:
Apologies for the confusion, but upon reviewing the example maze, I noticed that it is not solvable.
The maze has an isolated wall in the middle, blocking any path to the exit.
Here's the maze for reference:
* * * * * * *
* *
* * * * *
* * *
* * * * * *
* *
* * * * * * *
The maze as printed has neither an entrance nor an exit, and the wall the model points to in the middle would not obstruct a path.