The letter is not incredible. It contains only information that you provided! This is a sad use of an AI, making it into a text-to-text recognition system. It took longer to contrive this experiment and fix its omissions than it would have to write the same 2-paragraph letter yourself.
Holy shit, could I... write thank-you emails this way? :-O
I find thank-you notes horribly difficult to write for some reason, especially when the gift is not well-aimed (which is... usually, to be honest). I often end up skimming through 10-20 sample notes online before I can grind something out. Anything to make that process easier would be worth considering.
(This is honestly the first time I've seen a personally useful application for this tech.)
I know that's not the point you are after but that sounds like such a simple problem to solve.
Why are you thanking someone for a gift? Because they thought of you and went out of your way to get you something, and you appreciate them doing that.
A friend struggling with Christmas cards tried it and... yeah, it's great for that purpose. You still need to give it relevant context / facts, but it works.
I like that you can prompt "longer and with with more tangents".
You would probably enjoy Bureaucrats and How to Annoy Them[1] by R.T. Fishall, a nom de plume of Patrick Moore (of xylophone fame [2], maybe some astronomy too).
A few days ago, in anger, I asked ChatGPT for an email to Apple and I must admit that I finished the prompt with "The aim of the mail is to waste as much of reader's time as possible."
It was quite good, especially after a followup prompt "Not confusing enough. Make it more confusing.". But of course I didn't send it in the end.
I find it difficult to start letters and emails and end up with overwrought prose that's difficult to read.[1] Knowing this weakness of mine, I instructed ChatGPT to write a request to my doctor last week requesting that we meet to discuss adding a medication to my existing regimen.
The final request was simple, easy to read, and most importantly the process saved me the mental frustration I usually encounter when writing. I'm simply not the best writer as evidenced even now. That's a win in my book by any measure.
This is the bright side of the coin. The dark side is that it's by pushing ourselves that we improve. ChatGPT is going to result in a new equilibrium here...
Right, while I can hack out a few hundred words of passive-aggressive bureaucrat fodder in English in a few minutes, I know people for who, while they have excellent English, that's a huge drain and might take a very miserable hour. I am very, very grateful I've never had to navigate a non-Anglophone bureaucracy.
It is. I’m good at business letters, request emails and memos and have helped many coworkers and friends with them over the years. They are the experts in the domain of their request but wording it effectively isn’t easy for them. This is especially true when emotions are involved. I’ve been happy to provide help with the structure and general wording. But it’s better for them to have a tool that can get them going on a few seconds.
At the same time, expectations around correspondence will change when everyone can generate a decent document. There will still be opportunities to be recognized as an expert writer.
Not for everyone though. I find it a really useful use case. Sometimes I struggle with adding the "useless fluff expected in formal communication" that turns two simple facts into two paragraphs - and chatgpt is great at that.
Take us back to the year 2001 and this would have been regarded as witchcraft. Hell I think it's witchcraft now.
It's incredible how dismissive people are. I mean it's like looking at the wright brothers plane and saying, "I think this is a bad application of propellers."
I'm still completely amazed you can pluck the signal of cohesive language out from under the noise floor of large-corpus statistics. It's like something from Blindsight.
I thought code-division multiplexing's ability to pull a signal out from under a hundred dB of noise was magic, but this is far more unexpected, and I don't think you can honestly say anyone 10 years ago would have suggested this level of (linguistic) accuracy, speed and accessibility and not get laughed out of the room. "Ha ha, pull the other one, the novel-writing machines were in 1984, that's just an old science-fiction trope"[1].
[1]: It even used the word Ingsoc that's not in the prompt, and made the contextually correct connection to the Junior Anti-Sex League: https://pastebin.com/GQR9aRKT
Re [1] - ChatGPT can be The Versificator - the pulp generation device in 1984 - the versificator producing content for the proletariat such as tabloids (stories on sports, crime and astrology), films, low quality paperback novels, and lyrics. It seemed like an SF idea when I read the book, but now it is almost plausible, wow.
Yes, it's able to write stuff that we all leaned to write in primary school. What is the advancement in technology? I could write that email in 2 minutes, and probably take less time than the one required in instructing an AI to do so.
