GPT-3 the future for AI?

OpenAI the Californian non-profit company, started in 2015 with a goal to ensure that high performing AGI systems (artificial general intelligence systems) which could potentially outperform humans in most situations would be a benefit to human kind. (Tell that to those made redundant) Anyway, with backers like Elon Musk and Microsoft you know it must be true.

Of note is that it has now released the successor to GPT-2 , imaginatively called the GPT-3. A machine learning tool which has set social media ablaze with excitement (I’m stifling a yawn here).

In a manner reminiscent of Edison showing off his lightbulb where interested parties were shown the invention one at a time in a darkened booth and for a limited duration, so too was GPT-3 demonstrated to a few lucky people. The Iconoclast here wonders if they pulled off Edison’s trick, putting in a new lightbulb for every viewing (apparently their early bulbs had a life expectancy measured in minutes). But for those who have seen GPT-3 they claim how much more superior it is than GPT-2.

Apparently you can give it a missive, say some wordy legal letter and then show it a simplified version in plain English, so far so good. Next feed it a different legal document and then let it try and decode the legalese itself. Thus having been trained, the machine went on to do the task with great aplomb.

Seems it can also mimic the prose of some famous writers if given a few examples. But does it really understand what it’s doing in the way a human would? The jury is out on that one, however the head shed at Oxford University’s artificial intelligence research faculty Michael Wooldridge is dubious. He thinks it an interesting technical advance with a lot of potential uses but tons of data and a massive neural net, does not a human make.

So the Iconoclast here lives to write another day (until GPT-4 comes along maybe).

Leave a Comment