Next-gen ChatGPT sounds scary - it could be 'indistinguishable' from a human

ChatGPT has only just been upgraded to version 4, and we're already hearing leaks about the next-gen incarnation of the chatbot which we're told will be 'indistinguishable' from the experience of chatting to a human.

The next incarnation of ChatGPT, GPT-5, is being worked on by OpenAI and should be ready to roll towards the end of 2023.

According to a new report aired by BGR - citing Siqi Chen, a developer and entrepreneur on Twitter (see above) - GPT-5 might be set to reach a big milestone, namely achieving AGI or artificial general intelligence.

So, AGI - what exactly does that mean? Well, there's a problem here because there isn't a concrete definition of AGI, and depending on who you talk to or discuss AGI with, you're likely to end up with different ideas. Possibly quite varied ideas, for that matter.

Broadly, AGI means that the AI can be equated to a human-like level in terms of comprehension and completing tasks, hence the idea that it'll be difficult to distinguish talking to next-gen ChatGPT from chatting to a human on the net.

Chen puts forwards the argument that the "most primitive form" of AGI could be a self-prompting loop, with the AI being set an objective and then told that its first task is to create the next task, and so on. The AI then continues to generate those tasks, completing and reprioritizing them.

Whatever the case, the consensus, at least on Twitter, seems to be that the realization of at least a basic interpretation of AGI is imminent. Although Chen also clarifies that this is not the consensus at OpenAI itself, but that at least some staff members at the company believe that GPT-5 will reach this milestone.

Chen further theorizes that one of the reasons why we're apparently so close to AGI now is that nobody is even talking about the Turing test anymore.

For the unfamiliar, that's the test named after Alan Turing, which asserts that if you can speak to a computer and be fooled into thinking you're talking to a human, that device can be classified as artificially intelligent.

The reason why that's old hat is because GPT-4 has already hit that level (arguably GPT-3.5 could pass the Turing test fairly easily). At any rate, clearly the Turing test is an outmoded gauge of whether a computer is artificially intelligent, because GPT-4 most certainly isn't.

Within all this debate, it's worth remembering that AGI doesn't mean anything like the looming creation of a Skynet-style AI which will be the downfall (or savior, perhaps) of us all. It's a far more modest milestone than a true AI, or a super-intelligent entity that is still decades away (in theory), and able to far exceed the abilities of humans, not just match them.

That said, from the sound of things, GPT-5 will be capable of some seriously impressive tricks, because the current incarnation of ChatGPT can pull off some jaw-dropping stuff already - such as being able to turn a napkin sketch into a website instantly. ChatGPT just won't be governing the planet any time soon...