In this article, Cecilia Lo introduces some of the GenAI tools currently available, including text and multimodal tools, and explains their uses.
Cecilia Lo, Learning Technologist and Instructional Designer
In the previous step, you thought about some broad philosophical questions regarding generative AI. In this step, we return to a very practical approach. You will learn about, and experiment with, some of the types of generative AI tools available, and see what they are currently capable of (at the end of 2023).
Most of us have some experience with AI tools already, though we might not always realise it. For example, text autocompletion when you type; speech recognition in Amazon’s Alexa, Google Nest and Apple’s Siri; paraphrasing tools like Grammarly; and photo tagging suggestions on social media platforms.
AI in this context means a programme that uses machine learning.
In this step, we will look specifically at generative AI tools. In broad terms and as we saw earlier, generative AI tools are large machine learning models that take instructions (prompts) and create (or generate) credible-sounding novel content by applying a statistical model to material they are trained on.
Chatbots
First, let’s look at chatbots, which are tools that generate textual responses and have recently become very popular. They can also be called text-to-text AI tools. At the time of writing, the five most notable chatbots are:
- Anthropic’s Claude 2
- Google’s Bard
- Meta AI’s Hugging Face Llama 2 Chat
- Microsoft’s Bing Chat
- OpenAI’s ChatGPT.
Please note that these chatbots may not be available in all regions and territories; however, most of them are available in multiple languages.
Unless you have access to an Enterprise account from your institution, you will need to sign up for an account. You will need to agree to terms and conditions, and possibly to the use of your data for training purposes. Additionally, you need to be over 18 years of age to use Claude 2 and Bard. For Llama 2, Bing, and ChatGPT, you must be over 13, and those under 18 must have permission from their parent or legal guardian.
These chatbots can answer questions and write text, poems, tables, code and more based on your prompts. However, if you input the same prompt into different tools, you might find that they offer a different quality of answer. This is because of differences in training and fine-tuning.
Claude 2 and Bing Chat can read documents users have uploaded or opened. Bing Chat allows users to choose from different conversation styles (creative, balanced or precise) using a toggle button. While Bing Chat arguably provides the most explicit step-by-step on-screen guidance for users, all five provide examples to get users started.
Multimodal generative AIs
Chatbots are not the only type of generative AI tool. There are also tools that are trained with combinations of images, audio or videos, and that can generate multimedia as outputs.
You have probably already heard of text-to-image generators such as Adobe Firefly, OpenAI’s DALL-E 2, Midjourney AI, and Stability AI’s Stable Diffusion. These tools can:
- generate novel photorealistic images from text (eg the famous fake Pope Francis in puffer jacket image)
- replace specific elements within an existing image (eg replace a cat sitting on a sofa with a dog)
- create variations and different styles of an existing image.
Text-to-audio generators can create audio clips reading texts or scripts you provide in your own or a celebrity’s (eg former President Barack Obama’s) voice. Some of them can even use your own voice to read texts in a language that you don’t speak.
Both Google Labs and Meta AI are developing tools that can generate music based on text prompts.
Similarly, text-to-video generators offer a text-to-video editing function and can create video clips and animations directly from text description.
Furthermore, you can combine text-to-image, text-to-audio and text-to-video capabilities and create something that is very imaginative. Text-to-presentation tools, which are really tools that combine text-to-text and text-to-image capabilities, are already available.
You can certainly go further on your own. For example, Mermaid JS is a computer programming language that allows you to create diagrams from text. You can use a chatbot to generate the Mermaid code for your diagram and insert it into a diagramming tool like drawio.com to generate flowcharts, sequence diagrams, and mindmaps.
Another possibility is to use text generators to create a script for a video, then put the script through an audio generator to generate an audio clip. You can then feed the audio clip, along with images generated by an image generator, into a video generator to create the video.
Lastly, the capabilities of generative AI tools can be extended by adding plug-ins.
To use smartphones as an analogy, generative AI models are like the underlying operating system, and plugs-in are like the various apps (such as photo editing and note-taking apps) that you may install to allow you to do different things.
An example is AI-assisted search engines such as Bing AI and Google SGE (Search Generative Experience), which is only available in the US at the time of writing. They differ from traditional search engines in that they attempt to summarise search results for you.
Another example is AI-assisted meeting notetakers. There’s now even a brain-to-computer interface that can translate neural signals into text or words spoken by a synthetic voice. This tool can help stroke victims and paralysed patients communicate again.
More different types of tools are expected to become available as the field of AI develops, and updates to existing tools occur with incredible frequency.
Disclaimer: references to specific companies and tools do not indicate endorsem*nt.
Now that you have completed this step, you should be able to list several different types of generative AI tools and describe what they can do. In the next step, you will hear from some King’s students about their perspectives and attitudes towards generative AI tools.
Join the conversation
Now that you have learned about the types of generative AI tools that exist, it’s time to experience how they work and how they can be helpful.
Option 1
Go to two of the chatbot tools available in your location and ask them to do the same task. Then, compare the results. How do they respond differently? Do you prefer one over the other, and why?
In the Comments section below, post your task, the two responses you received and your evaluation of each.
Option 2
For a more advanced use of generative AI tools, ask one of the chatbots to write a very short story. Then use one of the text-to-media tools mentioned above and create an artefact (image, audio or video) representation of this short story. You may need to adjust your prompts a few times before you are happy with the result.
What do you think of the short story generative AI produced? How successfully does the text-to-media tool represent the story?
Share your reflection (and digital artefact) in the Comments section. Thinking about what you learned in the previous step, where do you think the creativity lies in these exercises?
© King’s College London
