Yes, GPT-4 is here. So what?

The king is dead, long live the king. While all eyes were still on GPT-3.5, Microsoft-backed OpenAI brought out its successor, GPT-4.

In fact, Microsoft’s new Bing has been run by GPT-4 for the last six weeks.

These leaps are longer and quicker than anyone – including the AI experts involved – had expected. As the Microsoft team said in a blog: “AI developments we had expected around 2033 would arrive in 2023 instead.”

While OpenAI says this is the smartest version of the chatbot so far, the makers say its output should be used with “great care” since the AI model still “makes reasoning errors” and has a tendency to “hallucinate.”

The description reminds me of a friend whose genius the rest of us admired, but occasionally, his solutions could be so off track that they could only be described as hallucinatory. So this AI is now more “human” than ever, ensuring that everything changes and nothing changes at the same time as we take GPT-4’s counsel with a spoonful of salt.

Interestingly, the latest edition of the chatbot follows instructions “very closely” and understands humor and sarcasm.

It’s already proven the breadth of its functionality. One company is using it to generate “one-click lawsuits” to sue robocallers for US$1,500, while someone else used it to look for security vulnerabilities in an Ethereum contract.

While the Duolingo owl has also been fortified with GPT-4, there’s also an experiment on to see how much money the large language model can make from US$100 through investments. Another saw it turn a hand-drawn sketch into a functional website.

The most fascinating experiment is one where GPT-4 outsources a CAPTCHA test to a human to get past guardrails.

But as something that learns from the world, Open AI says GPT-4 can also “generate potentially harmful content, such as advice on planning attacks or hate speech.”

By helping teach this AI language model the “rules,” one could also be teaching it to break them. (For more on this, check out the Waluigi Effect. Here’s some criticism of the theory as well.)

That said, it sure is a strange time for Open AI’s backer, Microsoft, to have let go of its AI ethics team.

This was published as a part of AI Odyssey, a section on generative AI developments featured in Tech in Asia’s emerging tech newsletter.