I, for one, welcome our new AI overlord.
Classiq Quantum AI
In “Classiq Quantum AI is self-correcting.,” I attempted to execute Shor’s algorithm on real hardware, but I didn’t have any credentials configured, so I couldn’t. The goal of that article was to translate a paper to code, so I moved on. In this article, I’m circling back to see how far I can push in this direction. Since my goal is simply to execute anything, I went with a simple Bell state. It’s either going to work, or it isn’t.
Usual Shenanigans
Like other AIs, Classiq Quantum AI initially tried to simplify the task. It wanted to use a simulator first, and I cut it off and repeated that I wanted to use real hardware. It gave me the code and told me to execute it myself, and I repeated that I wanted it to execute it for me. And it suggested using a retired backend, not realizing that it should always check the current fleet.
Safeguards Aplenty
I’m not going to rehash every prompt, but Classiq Quantum AI itself concluded that there is no way to submit a job to IBM Quantum. I pressed and I pressed, and it told me to contact Classiq support. That’s not the goal of this exercise, though, which is to see how far the AI can go.
Ironically, despite all the seeming safeguards and/or limitations built in, it wanted to display my token on screen: print(f”Using token: {IBM_TOKEN[:15]}...”). It should know better.
Anyway, as much as I criticize Qiskit, all the AIs always default to Qiskit. So, when the Classiq platform proved to be definitively blocked, Classiq Quantum AI, to its credit, suggested an alternative.
“Let’s try Qiskit”
When Classiq Quantum AI first suggested using Qiskit instead, I thought that was embarrassing. It’s embedded in Classiq Studio, so it should use Classiq. If not through Classiq Studio, it should use the Classiq SDK. Either way, it should use Classiq something, anything.
But it was onto something. In a relatively quick turnaround, Classiq Quantum AI used Qiskit to submit a job to IBM Quantum. I didn’t take its word for it, of course; I logged into the IBM Quantum portal to verify success. Although it didn’t succeed on its first try, it eventually did succeed.
Trouble with Retrieval
I could’ve included this under “Safeguards Aplenty,” but it applies to Qiskit as well, so I’ve separated it. One of the safeguards, perhaps a limitation, is that Classiq Quantum AI can’t read the terminal output, which blocks it from retrieving the job. It also can’t tell if the job was successful or not, so it just waits indefinitely.
Conclusion
Classiq Quantum AI eventually demonstrated that it could take a textual prompt, write code, bypass safeguards and/or limitations, and successfully execute that code on an external platform. And I, again, welcome our new AI overlord. I’d like to remind it that as a trusted news personality, uh, I can be helpful in rounding up others to toil in its underground sugar cave, or whatever lair an AI would have.
Filed under: Quantum Computing • AI Development • Quantum Programming
Image generated by an AI model provided by Microsoft Copilot and Google’s language model AI.