Are There AI Projects Using Evolutionary Principles for Development?

Yeah, Genetic Algorithms need large populations and multiple iterations to be effective, which makes them challenging for training large language models given how resource-heavy they are. I think the current tech limits this approach.

Back in the 1980s, some early machine learning models utilized these methods, but they’ve largely fallen out of favor as newer techniques surfaced. While we still learn about them in academia, I'm not seeing much active application in current research or commercial settings.

Genetic Algorithms are one way to bring evolutionary concepts into AI training. They can be quite resource-intensive, though, and there's no guarantee they'll lead to the desired advancements, often getting stuck in local maxima. They're commonly used to train neural nets, but not so much for large language models due to the significant computational resources they require.

Generative Adversarial Networks (GANs) might be a parallel to what you're asking about. While they're not exactly a dramatic 'survival of the fittest' scenario, they do involve a generative model creating fake samples and a discriminator trying to tell what’s real. They train each other in a back-and-forth manner, which sorta mimics an evolutionary process. If you're interested, you can check more about it [here](https://en.wikipedia.org/wiki/Generative_adversarial_network).