Google Gemini

In my recent exploration of language models (LLMs), I had the opportunity to try out Google Gemini and compare it with Microsoft Copilot.

First, I asked each LLM to generate text about human lives 6000 years ago. To my surprise, Copilot decided to focus on the roles of women during that era - an aspect I hadn’t explicitly requested in my query. In contrast, Google Gemini provided a more clean and polished answer, focusing on the general aspects of human life six thousand years ago without making assumptions about gender roles.

Secondly, I tested their translation capabilities by asking them to translate a English text into Russian. Here, Google Gemini outperformed Copilot (in my taste), providing accurate and fluent translations. This may be due to the fact that Copilot relies on Bing search results, which could introduce inconsistencies or inaccuracies.

It’s important to note that I still prefer using local LLMs over these cloud-based options, primarily because of their speed and privacy benefits. My go-to LLM remains Mistral, a neural network that caters specifically to my requirements without the need for an internet connection.