AI denies AI threat theory that 'I do not intend to exterminate humanity'

' GPT-3 ', the latest version of the language model 'GPT-2', which became famous as an AI that can generate high-precision sentences that are indistinguishable from those written by humans, said , ' I am human beings. I'm not going to make it extinct . '

A robot wrote this entire article. Are you scared yet, human? | Artificial intelligence (AI) | The Guardian

'I have no desire to wipe out humans,' robot writes in ominous op-ed --MarketWatch

OpenAI, which studies artificial intelligence (AI), released the further evolved language model 'GPT-3' in June 2020. The 'GPT-3' is the successor to the 'GPT-2', which temporarily postponed the publication of the dissertation because the development team said it was 'too dangerous' because it could easily and automatically generate text with too high precision. It is a model.

The latest version of the sentence generation language model 'GPT-3', which was regarded as 'too dangerous' because it produces too high-precision text, has been released --GIGAZINE

How the evolved 'GPT-3' spins words and generates sentences is explained in detail in the following article.

How does the ultra-high-precision language model 'GPT-3' spin words so that you can write a natural blog? --GIGAZINE

Using this GPT-3, the Guardian has created and published an editorial. In this article, GPT-3 said, ' The evolution of AI means the end of humanity ,' in response to Dr. Stephen Hawking 's words that AI is more dangerous than nuclear weapons. I am willing to sacrifice my existence for the sake of mankind. '

He went on to say, 'I'm not going to exterminate humanity. In fact, I'm not interested in doing any harm to you.' 'Eradication of humanity is totally uninteresting to me.' He states and denies the AI threat theory .

On the other hand, on the Internet, 'Guardians do not clearly explain that GPT-3 is a text generation tool rather than consciousness, and let GPT-3 generate sentences with the theme' Robots are useful for peace '. It is very irresponsible to publish this article without mentioning that it is. '

There are also criticisms of the Guardian, such as 'This article is a shameful cherry-picking of the Guardian, which makes naive readers more misleading about the facts about GPT-3.'

In addition, Guardian explained that the editorial that GPT-3 was generated was 'written by humans', and that he tried to write an editorial of about 500 words 'in a simple and concise language'. I wrote it. In addition, it is revealed that the editorial published in the Guardian is 'the Guardian's selection of the good parts of eight different sentences generated by GPT-3 into one editorial'.

in Software, Posted by logu_ii