I've been really frustrated with how AI handles writing tests. It seems like it has no real understanding of quality, just mimicking what it has learned without a grasp on what makes a good test. It often produces code that isn't coherent, especially when it comes to more complex testing scenarios like integration tests. The amount of prompt engineering needed just to get it to produce something decent is overwhelming. Is anyone else experiencing this? What are your thoughts?
5 Answers
You're spot on about AI lacking a real understanding of quality. It's basically trained to recognize patterns without knowing if what it produces is actually good. If you want reliable tests, you might just be better off writing them yourself from scratch.
Totally get what you're saying! The current AI tech just doesn't have that human touch, especially when it comes to grasping the logic behind what the tests are supposed to check. It often leads to incoherent test cases, which can be a real pain.
I've had success with tools like Copilot for PHP tests because it learns the patterns you use in your test cases well. But when it comes to JavaScript, especially with front-end testing—yikes! It really struggles to generate coherent tests for UI frameworks.
AI really acts like a fancy prediction machine—it's decent, but it can't replace human insight. You always need someone to double-check its output. It also relies heavily on the data it was trained on, which often consists of mediocre quality code.
Yeah, I’ve noticed that LLMs tend to default to the lowest common denominator, especially when it comes to tests. If you don't provide them with good examples, you might end up with something subpar. It's almost like you spend so much time teaching it that you could have just done it yourself.
Related Questions
Neural Network Simulation Tool
xAI Grok Token Calculator
DeepSeek Token Calculator
Google Gemini Token Calculator
Meta LLaMA Token Calculator
OpenAI Token Calculator