An AI Breaks the Writing Barrier
wsj.com
wsj.com
I wonder if GPT-4 will be able to take over the romance novel industry.
More likely carve a niche for romances made by AI.
Absolutely this. But the final form of this requires some input from the reader – front camera access and retina scanning presumably. It may evolve, without proper curtailing, (unidentifiable ads aside) to be a pretty powerful tool of motivation / pacification / manipulation.
>Eventually the content produced for particular readers will only be roughly literate, let alone clever or stylish.
Don't know about this, I have slightly more faith in people's ability of appreciation. Though I could be embarrassingly wrong.
I feel (a little hyperbolic allowances, if you will) it's in the books for an Infinite Jest type of entertainment to arise, be it in kindle or netflix form. That is, a piece of entertainment so excruciatingly entertaining, that people will literally and completely drop whatever it is they were doing before (or, you know, just miss a week of work for the sake of the most intimate and powerful writing they've ever experienced) and bask in the glories of their own personal god revealing truths of the universe to them.
Not everyone, in fact I dare do say the majority, don't want to read hyper-personalized content to the point that only we can read it, most of us don't even know what we want, and sometimes that doesn't even matter.
Hyper personalization isn't the end game - at least in my opinion. In fact history has shown that we valued precisely the opposite, collective experience with low enough resolution so we have room to fill the gaps ourselves. How can you relate to others what only you understand?
Or we will eventually cease to be humans?
I didn't understand the "optimism is cowardice" part. The true optimism is believing we will survive long enough to reach that point :P
I gave up on the story when it started to get romantic because I wasn't really in that kind of a mood at the time.
And to go further: a system that collects articles on the same topics and highlights semantic differences (facts, tone, framing, ...).
I'd love a ublock origin style plugin for identifying submarine advertising, astroturfing, etc. Could feed it press releases/advertisements and the like from major corporations.
Also, elaborating from / interpreting OP, to explore the parts of a given article which have absolute merit, and discovering patterns in effective language.
Obviously this won't be necessary knowledge for the sake of writing, though nonetheless interesting for linguists. But with the machines busying themselves with that, a good analysis and production could have a profound effect on spoken language. As in, you are what you read, and, unless you're still reading the classics, for a decent majority(?) that will necessitate an evolution in language. Not to mention the potential misuse by politicians and public speakers who will have a new level of delicate and gripping rhetoric.
For papers like the NYT, WSJ, or WP, every story with a byline involves a day's work at least. They read the feeds, sure. But if read carefully, you will notice (among other things) that almost every story will state that the journalist tried to get into contact with the major subjects of the story. It will either include their comment, or something like "... declined to answer/did not respond to inquiries/etc".
Then, there are the magazines, like the Atlantic or New Yorker. They make up a good chunk of what's generally considered "journalism", and rewriting wire stories just isn't something they do.
I don't like the way some words are used, and I complain about it too: https://coldewey.cc/2020/06/common-bugbears-of-modern-online...
That's our right as fellow users of the language. But unfortunately (such as with "begging the question" or "flammable") we're much too late to make a difference. AI is, I think one of those cases.
It's "an AI/ML/stats model" vs "an AI." "An artificial intelligence" has baggage that's hard to pin down, like it suggests it has a notion of self or sentience or something.
Seems legit: https://www.guinnessworldrecords.com/world-records/102759-la...
IMO, GANs were a much bigger and far more important step forward than GPT-2 ---> GPT-3.
Edit: after reading a few more comments, I concede that I may be ever so slightly underdoing it :)
It’s that itself it’s starting to cross the “Writing Barrier”.
The incremental gain is of little public interest, the social impacts of having an AI that can reliably produce content that is often indistinguishable from human content, now we’re potentially facing a profound change in how we write and in who’s writing we consume.
Maybe it’ll be an improvement
With regards to myself having a Smart Grammarly tool where I can write a page of dot pints and provide a few web links and then run a browser extension that turns it all into a coherent logical article. That would be brilliant.
Even more fake content -
The more SEO optimised, computer generate, algorithm ready content that is produced the faster we head into control-based dystopian context.
You spend a month and write up your best distilled, researched, coherent new thoughts on where we need to shift as a society.
Other parties jam together 1000 response articles drawing on established thought with the intent to deceive and your message is lost in the noise. “It’s all fake anyways”
An acceleration of our information downfall.
I wonder at what point it'll switch to "did a human write this" and I will click away when I realize it's inefficient monkey words.
Will the comments still be worth reading at that point?
[ this comment may or may not be GPT-3 generated ]
Please anyone?
Of course machines are capable of valuable discourse, because humans are biological machines, and humans are capable of valuable discourse.
Intelligent contributions are welcome whether they come from silicon-based or carbon-based life.
Judge the content, not the source.
Deciding on the legitimacy of an “opinion” comes down to meta ways of assessing; including the length of the text (HN rewards longer responses if they can maintain coherence and interest), the language style used (don’t make a technical mistake and use a technical term in the wrong context), the logic with which the argument is laid out, the emotional tone of the writer (No jokes on HN please).
All of these are now gameable leading to say for risk for example of state actors shifting our collective opinions on a huge range of topics.
I regularly make spelling mistakes in my posts and when I don’t correct them I loose points as I think the coherence and easy of reading goes down. GPT3 is unlikely to have the same issues.
And, by the way, journalism never was as matter-of-fact-only as everyone somehow assumes. Even you parents probably grew up in a Hunter S. Thompson-world. Your Grandparents had Hemingway who, despite the short sentences, didn't exactly crack the Shannon Limit in his war dispatches.
There have always been different styles. Just read the headlines only and you're almost back to the 19th century, when news-tickers actually ticked, and every tick cost a dollar, and a dollar is an hourly wage.
But I'd suggest actually understanding the purpose of narrative journalism and occasionally indulging in it. I've seen that complaint so often, and from a certain type of person almost exclusively, and I'm starting to wonder if it isn't some sort of psychological phenomenon, with guys (it's always men) fearing that anything less dry than a phone book rubs up on them, and their masculinity and rationality might suffer irreparable harm from exposure to yucky "feelings".
So...who's cynical here?
Financial news on lesser known public companies seems to be.
It's easy to generate a story based on some numbers about the stock to parameterize it.
Example (showing absolutely no understanding):
"Sorrento Therapeutics, Inc., belongs to Healthcare sector and Biotechnology industry. The company’s Market capitalization is $1.68B with the total Outstanding Shares of 32. On 03-09-2020 (Thursday), SRNE stock construct a change of -2.89 in a total of its share price and finished its trading at 7.06.
Profitability Ratios (ROE, ROA, ROI):
Looking into the profitability ratios of SRNE stock, an investor will find its ROE, ROA, ROI standing at -297.9%, -49.4% and -91.3%, respectively."