PolitePost turns even the crudest text into a neat email
There are those people who can even make a civilized garden out of the open sewer that the internet is at times. Like Anson Lai with PolitePost.
Are you one of those people who feel ashamed of their language every time they press Send? Then you will probably be very happy with PolitePost.
Hobby programmer Anson Lai put together this extremely user-friendly PolitePost app using a link with GPT-3 and a simple HTML frontend, which can be used to perform small miracles. You enter a raw text and the backend returns a neat and tidy email, with which you make a much better impression on a customer, a friend or your colleague.
Contents
Simple order with PolitePost
Here we give some examples. The first example is an order of eggs from a supplier.
You can copy this text from PolitePost into your mail client and press send. Not bad of course, but let’s put the system to the test a bit harder. For example with an unabashed swearing cannonade at an annoying neighbour. After all, it must be resistant to Dutch manners.
Angry letter to the neighbor
This is where the somewhat meager dataset of Dutch texts within GPT-3 takes its revenge. It took two attempts with PolitePost to generate this text. Only with the second attempt and the addition of “In Dutch:” at the beginning, a good text comes out. Probably also because Lai has probably primed the AI with English text, things are less smooth in Dutch.
Raunchy tweets from politicians through the wringer
Of course it is a mortal sin to mention politics, but I couldn’t resist the temptation to put some tweets through this filter from some politicians who have a reputation for foul language.
Remarkable. The AI rewrites Wilders’ tweet as Rutte’s excuse. The rather fierce, compelling tone of Wilders’ tweet probably drives the AI to this weighting. This is an emergent phenomenon, in plain language: a spontaneously arising effect.
This is very surprising and unexpected. Would a similar effect also work on humans, you wonder. Occasionally you see emails about “hypnotic writing”. Would its effect be based on a similar mechanism? Could you test the hypnotic power of a text in that way? To level the political playing field a bit, a swipe from D66 member Tjeerd de Groot.
The AI misses the mark here, probably because the Boer Burger Movement did not exist yet in 2019, when the last texts were added to GPT-3. As a result, the farmers are designated as perpetrators, where De Groot wants to portray them as victims of deception by BBB.
Conclusion PolitePost
For simple emails this is a great tool and for emails that are about things that existed before 2019. It is perhaps less suitable for texts with a high level of abstraction. And always read the text before sending it. But apart from that, it is a tool that you can enjoy a lot.
Would you like to support Anson Lai with a small contribution of, for example, €2 or €10? Then you can do so via the donation page. API Requests via OpenAI are not free.
