There are also now credible alternatives to ChatGPT.

These include Anthropic’s Claude, Google’s Bard and Microsoft's Copilot and many other products based on the same Large Language Models that give generative AI tools their power.

Some are free, but many of the most powerful ones require payment.

Considerable advances are also being made in image and audio generation. What is clear is that generative AI is here to stay, and is already being used by people to make real and meaningful contributions to assist with their work.

The new era of generative AI introduced itself from a chatbot.

But this simple interface belies a wealth of potential the extent of which has still not been fully explored.

Among the unexpected uses that have emerged, over and above simply generating text, are extracting information from text (for instance, finding people mentioned in text); presenting information in tables and/or structured lists; asking it to generate and/or correct computer code in a variety of languages; generating multiple choice questions about text; explaining and analysing abstract concepts including metaphors; and answering questions about well-known facts (including the contents of some books).

Despite advances on many fronts, these tools remain unpredictable, and in an academic setting their output must come with a healthy caveat. 

Large Language Models (LLMs) are trained to generate text that is likely to occur, and so will often generate plausible but entirely fictional facts – because of their probabilistic nature, these will be different every time.

In fact, even the original announcement of ChatGPT included the warning that it sometimes writes ‘plausible-sounding but incorrect or nonsensical answers.’

For example, non-existent links, titles of books or papers, made up biographical details about known people – these statements will often be seamlessly embedded within perfectly factual statements.