An article written by 'GPT-3' that was regarded as dangerous as 'making too high-precision text' emerged as the first place on social news sites

A university student who thought that 'GPT-3 has the power to change our 'writing'' opened a blog that let GPT-3 write articles for demonstration, and it was 26,000 in two weeks. It seems that people visited and temporarily ranked No. 1 on the social news site

Hacker News . Even after so many eyes, only a few people doubted that it was due to GPT-3.

My GPT-3 Blog Got 26 Thousand Visitors in 2 Weeks-Excavations

In July 2020, engineer Manuel Alaos published a blog post on GPT-3, which was a big topic.

Why is the blog post that 'GPT-3 has a destructive potential since Bitcoin' so big? -GIGAZINE

Liam Porr, a computer science student at the University of California, Berkeley, took notice of Manuel's attempt in the second half of the article, saying 'it was all GPT-3 generated so far'. I created a blog that does not reveal that it is GPT-3 generation. Of the articles posted on the blog, Porr prepared the title, introduction, and photos, and GPT-3 was in charge of writing the text.


When Mr. Porr aggregated the data for 2 weeks, the number of blog visitors was 26,000, the number of subscribers was 60, and it was pointed out directly that 'It is a sentence generated by GPT-3' 1 There were only people. Regarding this, Porr accepts that people who argue about the sentences generated by GPT-3 as 'how incoherent or irrational' seem to like another matter whether they like or dislike the sentence. ..

This article was especially popular.

Feeling unproductive? Maybe you should stop overthinking.-- Adolos

Hacker News, a social news site, has over 70 comments.

Feeling unproductive? Maybe you should stop overthinking | Hacker News

According to Porr, he was temporarily at the top of Hacker News. Hacker News is a mechanism that allows users to vote on topics that they should 'go further up', and it can be said that it has collected that much vote.

Although there were a few doubts about GPT-3 generation within this Hacker News topic, other users pointed out that it was contrary to Hacker News's policy as a community, and the display priority of comments was lowered. It seems that few people noticed the comment because they received 'downvote'.

In response to the results of this demonstration, Mr. Porr felt that 'Using GPT-3, it will be possible for one excellent writer to replace the content production team of multiple people'. That means. There are various problems such as illogical parts in the text generated by GPT-3, blurring topics, repeating the same story, etc. It was readable without any problems if I deleted it and added a conclusion. In the proof-of-concept, for the conclusive evidence that Bob Dylan's citation was mistaken for Ralph Wald Emerson , or there was a simple grammatical mistake, it was apparently generated by GPT-3 It seems that it has been fixed, but not the others.

Of course, it is difficult for GPT-3 to generate fact-based reports, but Porr feels the possibility as an auxiliary tool, and he is also making specific trial calculations.

For example, on a news site with 400 authors, suppose GPT-3 improves efficiency by 50%. We also assume that only half of the authors will benefit from GPT-3. In this case, if GPT-3 is used, the work equivalent to 200 people who do not use GPT-3 can be done by 133 people who use GPT-3, which can reduce 67 people. Since the average monthly salary of a writer in the United States is $4,200 (about 440,000 yen), the labor cost that can be reduced will be about $3.4 million (about 360 million yen) per year. Porr predicts that this number will be higher unless GPT-3 content production fails significantly.

The appearance of the GPT-3 is like the appearance of McCormick Reaper (a reaper invented by Cyrus McCormick) in agriculture, Porr said. With the emergence of the reaper, it may be unnecessary to 'make sentences from scratch' just as the manual work with a sickle becomes unnecessary.

in Posted by logc_nt