The Arrival of GPT-4

 

 

 

 

 

 

 

 

 

 

The flurry, or rather avalanche, of announcements of new variants of generative AI have brought a storm of hype and fright. 

 

OpenAI’s ChatGPT already appeared to be a game-changer, but now this week’s new version, GPT-4, is another leap ahead. GPT-4 can generate enough text to write a book, code in every computer language, and – most remarkably – “understand” images. 

 

This is the “next wave” of technological change. Now generative AI has moved potential progress up a gear or two. 

 

Artificial intelligence in the form of large language models (LLMs) – such as ChatGPT – are an accelerant that operate at such scale and speed that they can appear to do whatever you prompt them to do. And the more that we use them and feed them data and questions, the faster they learn to predict outcomes. 

 

A million start-ups are already claiming to use it helps them create new products that will revolutionise everything from legal administration to share dealing, gaming to medical diagnosis. 

 

Of course, a lot of this is marketing froth. As with all tech breakthroughs, there is always a hype cycle and unexpected good and bad consequences. But I’m sure it’s going to alter our lives one way or another. 

 

Artificial intelligence such as machine-learning, automation or natural language processing is already part of our world. For example, when you search online you are using machine-learning-driven algorithms trained on vast datasets to give you what you are looking for. Now the pace of change is picking up. In 2021 alone, global private corporate investment in AI doubled, and I expect the generative AI breakthroughs to double that again. 

 

Now take a breath. I don’t recommend that anyone uses ChatGPT or GPT-4 to create anything right now – at least not something that will be used without a human checking to make sure that it is accurate, reliable and efficient, and does no harm. AI is not about the total automation of content production from start to finish: it is about augmentation to give professionals and creatives the tools to work faster, freeing them up to spend more time on what humans do best. 

 

We know that there are some real extra risks in using generative AI. It has “hallucinations” where it makes things up. It sometimes creates harmful content. And it will certainly be used to spread disinformation or to invade privacy. People have already used it to create new ways to hack computers, for example. You might want to use it to create a wonderful new video game, but what if some arch-villain uses it to create a deadly virus? 

 

We recognise those risks because we can see its flaws when we try out these prototypes that the technology companies have made publicly available. You can have a lot of fun getting it to write poems or songs or create surreal images. Ask it a straight question, and you usually get a sensible safe answer. Ask it a stupid or complex question, and it will struggle. And surely, AI boffins will be super delighted because this all helps refine their programming. 

 

But who decides in the end what risks are acceptable or what we should do about them is a moot question. 

 

To Charlie Beckett (left), a professor in the Media and Communications Department at the LSE, it is too late to put this technology “back in the box”. It has too much potential for helping humans meet the global challenges we face. 

 

He adds, it is vital that we have an open debate about the ethical, economic, political and social impact of all forms of AI. And we should take the time and effort to think carefully about the best ways that it can be used positively.