Now, let me just say this upfront: if you’re looking for a comparison that involves fancy algorithms or complex math equations, you might want to look elsewhere.

First off, Koala. This little guy is trained by fine-tuning Metas LLaMA on dialogue data gathered from the web. In other words, he’s a chatbot that can respond to your queries based on what he’s learned from conversations online. Now, we know what you might be thinking “But isn’t Koala just another AI model? What makes him special?” Well, my friend, let me tell you something: Koala is not just any ordinary AI model. He’s a chatbot that can execute instructions! That’s right, You can ask him to do things for you like brewing coffee or folding laundry (although we don’t recommend the latter).

But enough about Koala, Alpaca. This guy is also trained on dialogue data gathered from the web, but he’s not as fancy as Koala. He can only respond to your queries and provide you with information no instructions for him! Now, some of you might be thinking: “But isn’t that what Google is for?” Well, my friend, let me tell you something: Alpaca doesn’t have access to the internet like Google does. So if you ask him a question about current events or sports scores, he won’t be able to provide you with up-to-date information.

Now, let’s compare these two chat-based systems in terms of performance. According to a recent human evaluation conducted by researchers at the Berkeley Artificial Intelligence Research Blog (BAIR), Koala performed better than Alpaca on their proposed test set which consisted of real user queries posted online. In fact, Koala was rated as better than Alpaca in nearly half the cases and either exceeded or tied Alpaca in 70% of the cases!

But wait a minute what about the Alpaca test set used by Stanford’s research team? Well, according to BAIR’s evaluation, Koala-All (which is trained on both ChatGPT distillation data and open-source data) exhibited comparable performance to Alpaca on this in-distribution dataset. However, when it comes to real user queries posted online, Koala-All was rated as better than Alpaca in nearly half the cases!

While both of these models are trained on dialogue data gathered from the web, Koala has an edge over Alpaca when it comes to executing instructions based on real user queries posted online. But hey, that’s just our opinion you might have a different one!

Until next time, keep learning and exploring the world of AI!