[calderknight]

Mystery solved! The answer to your OP is that the reason anyone can trust ChatGPT for code is that they use a much better model than the one you're using! GPT-3.5 is ancient and way behind GPT-4. In fact, there are now tens of organisations who have developed model classes way ahead of GPT-3.5.

[chrisjj]

> the reason anyone can trust ChatGPT for code is that they use a much better model than the one you're using! GPT-3.5 is ancient and way behind GPT-4.

Testing that theory now using Bing Copilot GPT-4.

Failed.

unmatched close parenthesis Line: 11, Column: 11, File-Position: 408

Input is:

  Translate this to Common Lisp.

  def solve(graph):
    a = 0
    while a < len(graph):
      for b in range(len(graph)):
        if b > a and graph[a].intersection(graph[b]):
          graph[b] = graph[a] | graph[b]
          graph.pop(a);a-=1
          break
      a+=1
    return graph

  a = [{1, 2}, {4, 5}, {1, 3}, {1, 4}]

  print((solve( a )))

Output is:

  (defun solve (graph)
    (let ((a 0))
      (loop while (< a (length graph)) do
        (loop for b from 0 below (length graph) do
    (when (and (> b a) (intersection (nth a graph) (nth b graph)))
      (setf (nth b graph) (union (nth a graph) (nth b graph)))
      (setf graph (remove-if (lambda (x) (equal x (nth a graph))) graph))
      (decf a)
      (return))))
        (incf a))
      graph))

  (let ((a (list (list 1 2) (list 4 5) (list 1 3) (list 1 4))))
    (format t "~a~%" (solve a)))

[calderknight]

Yeah, if going that route the answer to the OP is "most people aren't using it for Lisp".