Examples of Generative AIs: source @ethersiim
There is commentary about how AI is going to transform every area of our work and home lives. The software testing space is no exception. But to understand the impact AI could have on our world, it’s worth going back to basics to understand the generative AI concept that underpins ChatGPT and other tools like it.
ChatGPT is perhaps the most famous example of a generative AI (the GPT of its name stands for Generative Pretrained Transformer).
Generative AIs are trained on Web 2.0 content such as text, images and audio. They are fine-tuned with both supervised and reinforcement learning techniques and can then generate new plausible content.
ChatGPT is one of the largest trained language models because it’s underpinned by 175 billion machine learning parameters.
However, you get a sense of how far we still have to go in this field when you consider that its inputs are dwarfed by the Megatron-Turing Natural Language Generation model (from NVidia and Microsoft), which has the largest dense neural network with 530 billion parameters, and PaLM: Scaling Language Modeling with Pathways, which was trained on 540 billion inputs.
The wealth of inputs means there are few things generative AIs such as ChatGPT can’t do. The fact it can write code unlocks interesting opportunities in the software development process.
I’ve used ChatGPT to undertake the following:
The most exciting possibilities come with the volume at which AI can execute, especially in the Operate phase. In theory, you needn’t have any tests running. Instead, you could have real users acting as the testers, finding the bugs, and letting the generative AI fix them.
But there are challenges with a scenario in which we rely too heavily – or perhaps too unthinkingly – on the capabilities of generative AI.
In the mainstream media, all the talk has been of how ChatGPT has passed advanced examinations in medicine, law, and finance. You’ve probably seen in our world that it passed the ISTQB® Foundation Level Sample Exam for ISTQB CTFL Certification. So does that mean ChatGPT is a doctor, a lawyer, or a certified tester?
You can spot the problems with this premise when we start to think about how generative AI could become embedded in our everyday lives. What if a generative AI made a faulty medical decision that resulted in a death? Or if one was found to exhibit racial or gender bias?
The ramifications are already being considered by the EU. The proposed EU AI Act proposes that infringements could lead to fines of up to €30,000,000 or up to 6% of total worldwide annual turnover, whichever is higher.
Even on a day-to-day level, there are technical complexities. How do you test AI technologies?
Randy Hesse, Product Solutions Architect at Keysight's Eggplant outlines how this approach reflects the way the use of generative AI is going. Our AI-powered testing platform simulates user behavior, generates test cases, and evaluates the performance of the AI system or model.
This approach is perhaps best summed up with what’s fast becoming a mantra in this space – AI won’t take your job but someone using AI will. Because for me, generative AIs give us augmented intelligence or augmented testing capabilities, helping us with day-to-day tasks, whether it’s exploratory testing, automated testing, or performance security.
For example:
联系客服