[jll29]

We are just building a chatbot to assist students that study (that study AI itself, to be precise).

We could not "just use ChatGPT" because the answers have to be correct, or students could take legal action against the university in situations where they'd give an answer in an exam that the chatbot gave them and that might be wrong. It's exciting for sure to compare e.g. ChaptGPT with other approaches, for a set of controlled benchmark questions such as:

  -- snip --
  Q: What is the A* algorithm?

  A: The A* algorithm is a popular heuristic search algorithm used in pathfinding and graph traversal. It is an extension of Dijkstra's algorithm, but uses a heuristic function to guide the search towards the goal node, which can make it more efficient than Dijkstra's algorithm in certain situations.
  The algorithm works by maintaining a priority queue of nodes to be explored, where the priority of each node is the sum of its actual cost from the starting node (g-score) and an estimate of the cost to reach the goal node (h-score), based on a heuristic function. The heuristic function must be admissible, meaning it never overestimates the actual cost to the goal node.
  At each step, the algorithm selects the node with the lowest priority from the queue and expands it, generating its neighboring nodes and computing their priority scores. The algorithm then adds these nodes to the queue if they have not been explored before or if a better path to them has been found.
  If the goal node is reached, the algorithm terminates and returns the path from the starting node to the goal node. If the queue becomes empty before the goal node is reached, then there is no path from the starting node to the goal node.
  Overall, the A* algorithm is a powerful and widely used algorithm that has many applications in fields such as robotics, video games, and logistics.
  -- snip --

[aimbivalent]

This is variation on the common "it has to be perfect" argument. I don't really buy it.

Right now, GPT 4 would earn a top 10% SAT score, implying that it can give very high quality answers on a range of scholarly topic. This output is basically free.

Nobody that's beholden to even mild economic pressures is going to pay for an expensively constructed 100% solution if they can have the 99% solution for free.