For the past two years, the co-pilot, GitHub’s Coding Wizard KI, has been the researcher at the Anlass Genommen Hat, an in-depth study of code theme quality. It is time that the Code has been created by KI-Hilfe, more and more unit tests have been passed and new tests have been carried out. The reviewer felt like he was an experienced interpreter and interpreter.

Anzeige

Hat GitHub 202 Python-Entwicklern Thinking about time management services past, an API endpoint for a web server is designed to create a restaurant management system. 104 of them used GitHub Copilot, 98 have five KI assistants. The tester presents all unit testing examples, with the full testing features and testing scopes available to a KI enterprise at: 60.8 percent of the program best for all testing with Copilot, at 39.2 percent without KI. -Hilfe.

25 ausgewählte Entwickler, deren Code alle zehn Tests bestanden hatte, sollten anschließend blinde, anonymisierte Reviews der Program anfertigen, wobei jedes Programm dabei insgesamt zehnmal geprüft wurde. You can read more in these Schritt nicht mehr Funktions-, sondern qualitative Mängel, die Konsistenz oder Lesbarkeit betreffend: inkonsistente Benennung, unklare Identifier, zu lange Zeilen, zu verschachtelte Schleifen, fehlende Kommentare, wiederholte Ausdrücke (do not repeat yourself, DRY) or unsaubere Functionssafety.

Even if it’s your gut, you don’t have to worry about it. I am Schnitt fanden die Reviewer 4.63 Fehler bei Programn mit KI-Hilfe e 5.35 Fehler bei solchen ohne. Also the Zahl der Codezeilen pro Fehler war ohne KI-Hilfe größer: 18.2 Zeilen zu 16 Zeilen. Broader studies should include the aufgeblähten code and an applied version of the DRY principles. Currently being studied by Annahme, the quality of the information is in question.

Also, the Reviewer is still on the same page, if it is less, less, less, less, less and less important, the Code is already in use. There is an advantage of 3 and 5 points for the co-driver, the study does not take place, as the stand wants.

The studies made a click on the project commitments, with the KI-Hilfe häufiger and small ausfielen.

The abstract shows the study for: “Our hypothesis is, although the hypothesis is good, a working code for the machine, which is more suitable for quality experts.”

(WHO)