Quantum computing has been promising to change the world for years, but always from the distance of laboratories. Now, a real experiment has just moved that promise several steps forward, and the results surprised even the researchers themselves.

A group of scientists managed to improve the performance of an artificial intelligence model by connecting it to an IBM superconducting quantum computer. Not only was the system more responsive than before, it was also able to correctly answer questions that the original AI model simply couldn't solve. That, in the world of technology, is no small feat.

The quantum machine that taught an AI to think better

The experiment was developed by researchers from Multiverse Computing, and to carry it out they used the IBM Quantum System Two, a 156-qubit quantum processor, along with a framework called QIML. But here comes the interesting thing because the goal was not to build a quantum AI from scratch, but something much more intelligent from a practical point of view.

The scientists developed a hybrid system that combines a quantum generative model with a classical predictor. The quantum part generates what they call a Q-Prior, an advanced mathematical representation capable of detecting complex relationships and tiny patterns within the training data. Without that quantum component, the model's predictions tended to degrade and lose stability over time. With it, the answers became more coherent, more reliable and, in some cases, directly more correct.

The model that was improved with this technique was none other than Llama 3.1 8B, the 8 billion parameter language model developed by Meta. What is impressive is that the quantum intervention only added about 6,000 additional parameters, which is equivalent to just 0.000075% of the total model, and still achieved a 1.4% complexity reduction.

Answered questions that normal AI couldn't answer

Here's the part that really surprised the researchers. The hybrid system was able to correct errors that the original model made in specific scientific questions.

One of the most striking cases has to do with astronomy. When the Llama 3.1 standard was asked which Jovian planets have rings, it responded that only Saturn does. A classic mistake, quite common even among people. But the version optimized with quantum computing correctly identified that all the gas giants in the Solar System have ring systems, including Jupiter, Uranus and Neptune.

Another example came from the field of genetics. The base model offered an incorrect explanation of Hardy-Weinberg equilibrium, a fundamental concept in population biology. The improved version, on the other hand, adequately answered that gene flow increases genetic homogeneity between populations, demonstrating that quantum intervention not only improves numerical precision, but also the quality of scientific reasoning.

And the numbers back this up. In some benchmarks, the system increased precision up to 17.25% and raised spectral resolution up to 2.936% compared to comparable classical methods. Those figures speak for themselves.

An experiment that can take AI and quantum computing to another level

Borja Aizpurua, senior scientist at Multiverse Computing, was clear in explaining the project's approach. The goal is not to replace current models with fully quantum systems, but rather to leverage specific quantum capabilities to improve efficiency and representation of complex data. It is a pragmatic vision, and that makes it much more powerful.

In addition, there is a technical detail that makes this advance especially relevant for the future. The quantum computer was only involved during the training phase, not during model execution. That means that, once trained, the system can run on conventional classical infrastructure, making it viable outside of laboratory environments and potentially scalable in the real world.

This also points to a solution for one of the AI ​​industry's biggest headaches today. Language models have been growing uncontrollably in size and computational cost for years. This experiment shows that perhaps you don't always have to make them bigger to make them better. Sometimes, just making them smarter in a different way is enough.

For years, artificial intelligence and quantum computing advanced as two parallel worlds that rarely touched each other. This experiment is a clear sign that that distance is shortening, and that the real revolution could come not when one of the two technologies dominates the other, but when both learn to work together.