In the majority of expressions I’ve tried, AI provides astonishingly good explanations, sometimes surfacing subtleties I hadn’t noticed. On the other hand, it can occasionally produce some very wrong explanations. It does seem to be getting better with time, which is to be expected.
As I consider if and how to implement this kind of “AI assistant” in QSDA, I have some questions for the community.
Would you find a feature like this useful? Keep in mind that the accuracy will never be 100%.
In your role do you encounter unfamiliar expressions? For example, you are a consultant or team member tasked with working on existing apps. Or do you mostly work with simple expressions or expressions you wrote?
The expression text would be submitted to and processed by the OpenAI/ChatGPT server using an internet connection.
Would processing data like this be allowed by your organization’s policies?
Does your organization currently have or is contemplating policies regarding using AI?
This feature would require an OpenAI account – and a payment method.
Does your organization currently have an OpenAI account?
Let me know if you want to test drive it and see how it works with your apps. You will need to be able to establish an OpenAI account and allow connection between your QSDA server and OpenAI.com.
LLMs like ChatGPT can indeed be of great help when dealing with someone else’s work. I am thinking about when you need to take over an application or when you need to review the application before moving it to quality/production, or simply when facing performance issues.
The actual next step would be for these tools to suggest an improved expression that would perform faster
ChatGPT is not allowed in my organization. This is an open model and it has a memory, we cannot feed ChatGPT with company internal knowledge, even if this is only about expressions, not data points.
OpenAI is most probably considering private instances of ChatGPT, that could be an option in the future. The pace at which the Open Source community is building on top of existing models is quite impressive. (https://twitter.com/jelleprins/status/1654197249981796358)
You can find tons of models on GitHub that do well for specific jobs. I don’t think anyone has created a specialized model for Qlik Sense and I am not saying you should, but someone will most likely pick up the opportunity at some point. Such models could run at no extra cost on the QSDA server, since they run well on a laptop and can be trained in a matter of a few hours.
I can be wrong, but I believe the future stands with these specialized models that every product/company will build and feed with internal/specialized knowledge to give them an edge on the competition. ChatGPT does not know your competition, what’s driving your sales, it can make educated guesses, but what if you can feed it with internal reports that will teach its model how a product launch or supplier shortage can impact manufacturing capacity and n months later the sales revenue in a specific market… That would be a business analyst teaming up with AI.
Bottom line, OpenAI is a quick win, which, if allowed by the organization, can help a bit the unexperienced to medium experienced Qlik Sense developers.
For those interested, they can already give it a try for free with a simple prompt on https://chat.openai.com, or if they have an OpenAI account try it out with QSDA.
Maybe you have already optimized the costs by looking at the available models on OpenAI API