On 6 December 2023, Google announced Gemini, a family of AI models in three sizes: Ultra for the most complex tasks, Pro for a wide range of tasks and Nano for use directly on phones. Google described it as "natively multimodal".
What is multimodal?
Earlier systems often handled one type of data: text or images or sound. A multimodal model is trained to work with several types together. You might show it a photograph of a maths problem and ask for an explanation, or give it a chart and ask what it shows.
The wider picture
By the end of 2023, there were many strong AI models: OpenAI's GPT-4, Google's Gemini, Anthropic's Claude, Meta's open Llama models and others. They competed on accuracy, speed, price and safety. That competition is good for users, but it also means the tools and their abilities keep changing.
How to cope with so much change
- Learn the ideas, not only the buttons. Prompts, checking and understanding data stay useful whichever tool you use.
- Try more than one. Different tools suit different jobs.
- Keep a human in charge. The person using the tool is responsible for the result.
We keep an eye on new tools so that our lessons stay current without chasing every headline.
