Writing for Snobs
russelldavies.typepad.com
russelldavies.typepad.com
And nobody reads marketing brochures anyway. I hope they don't, and I pity them otherwise.
Who I don't pity though are lawyers, and if GPT-3 can make lawyers as miserable, as normal people are when forced to deal with lawyers: good. I mean, I don't really believe anything can eliminate this pesky profession (which depends on laws & litigation process being incomprehensible to ordinary citizens), but I can at least entertain myself with this thought for a while.
Of course, it's fun to chuckle at the thought of a lawyer having to read through the most insane document of all time and muttering profanities to himself the entire time.
What GPT-3 can be good at though is making a huge amounts of text cheaply and quickly. Making up a tons of bullshit to present in court (or putting the other side in a position when they have to) is a way to make litigation last forever, i.e. until the side with less money is forced to give up on a case they could have otherwise won. You can win by virtue of having more lawyers to read and type stuff.
And that way to use GPT-3 based stuff I imagine to be much closer shot. A company with 2 competent lawyers will be able to generate virtually same amount of semi-coherent bullshit as a company with a 100 of lawyers. You can produce an amount of stuff in a single night that is humanly impossible to read, but still if kinda meaningful stuff with some content kinda "relevant" to the matter at hand. At some point it will stop making any sense, the judge won't read this shit, nobody will. You will have to somehow forbid to make stuff more complicated than it must be. What is currently normal will fall apart.
As I said, I don't actually believe it will do any good: people will still find a way to make it way too complicated for normal humans, but acceptable for lawyers. But until they agree on a new norm, it will be painful for everybody involved.
For example, I've had to read quite a few 1,000+ page legal documents (Companies Act 2006 being one example) and that document gets referenced all the time. I can imagine tools like GPT-3 being used to crank things up to 11 to ensure that no user ever really reads or understands the TOS/EULA/whatever they're reading. Not that most people do anyway.
> too complicated for normal humans, but acceptable for lawyers.
I can totally imagine someone building a tool to parse generated garbage legalese into plain language bullet points and adding the proper visual references/links and document structure so that a human can read it. Perhaps it already exists and I'm just not aware of it.
The only reason GPT-3 sometimes gets facts right is that it was trained on Wikipedia. It will be as inconsistent as its input.
Notably, this has been true for visual content from the beginning, compare the original "TRON" movie. It had been especially true in times, when smooth gradients and perfect lines had been a luxury, which were hard to achieve in analog media technology, and became the tell-tale signature of computer generated visual content. After an intermezzo in the uncanny valley and with embraceable pixels, we seem to be right back where we started. – It's somewhat logical that what is true for visual content should also be true for textual content.
[Edit] We may add a definition: "Interesting" is the artful deviation from the smooth and perfect.
hard to achieve as a practical effect, but still cheaper than digital.
See also the expressed scanlines in the "faked" 80's of the first video in https://news.ycombinator.com/item?id=24398456 .
As for exaggerated curvature, surely we've all run across https://github.com/Swordfish90/cool-retro-term by now?
The goal of photography is to look like CGI.
The goal of CGI is to look like photography.
The topic of the article wasn't all that interesting to me, but I thought this quip was interesting.
Does anyone have any references that expand on the merit of it?
Or thoughts on the idea in general?
Even black t-shirts are not all the same if you know where to look.
Hell, if we both have a Porsche 911, but I share it on instagram as my wonderful car and you just never mention it because a Porsche 911 is a normal car to you ... which of us is higher class?
Then there are shibboleths. We use this in professional contexts. If you’re a Fortune500 person and I say I will consult my lawyer, you know I’m a schmuck and a small vendor. Otherwise I’d have a department and would say I will “talk to legal”
Edit: the point was that shallow things easily copied with minimal investment make bad signals.
If you’re not part of the in crowd you’ll never know what glaringly obvious signal makes you stand out.
For instance it is an old people’s saying in Japan to look at someone’s shoes to know if they are rich/value good objects (the basis of that was that poor/undiscerning people would have poor shoes maintenance, or plain bad shoes).
Or you would tell someone’s rank in a company by the amount of stuff he (was about men, yes) was carrying every morning (basis was higher ranked people would have less manual roles)
Of course these saying become pointless and people find other stupid stuff to look at to rank other people on a scale. But yeah, in my short experience it’s a thing people train to do, and is valued as useful skill.
Also as people tend to dress rather uniformly, so you need a decent amount of effort to stand out without being flashy, which means trying to prove good taste with exotic and expensive but plain looking goods, that you justify owning by touting their good quality and not their high price.
That said, I think there is a correlation between the amount of social signalling one must do to be successful in a society and the amount of wealth that that society allocates to those that are capable of navigating organizations. It's one of the reasons I think the tech sector is mostly jeans and sneakers.
Fancy shoes don't mean you're capable of knowing what a gin index is or when to use it.
While I have many qualms with modern tech, this is not one of them.
Ideally your rain jacket is appropriate for weeks-long alpine or warmer-weather arctic expeditions, and costs at least $250. It should pack down unbelievably small so you can excuse/brag-about the expense as an expression of "minimalism". You wear it on your $1000 bike for your 2km commute to the coffee shop to work on your MacBook. Shoes should be ready for many kilometers of hard, high-performance trail running, and should have been priced to match. Or else rock climbing or bouldering. No activity is more hip-tech-culture-approved and laudable than rock climbing, at least for now (the trends come and go).
There's that stuff, then there's the hipster-mimicking tech-fashion sect in $200 Japanese selvedge jeans and $300ish Red Wing Heritage workboots. There's plenty of overlap and mixing of the styles, though.
Don't forget the right hoodie. If the aglets aren't metal and the zipper's not a heavy, smooth-but-crunchy-feeling YKK, you've screwed up.
If in doubt, just shop REI and Patagonia. Little they stock will be too far off the mark.
I am not joking.
"Look at my fucking red trousers" blog -> http://lookatmyfuckingredtrousers.blogspot.com/
followed by a piece in The Guardian "in defense of red trousers" which cites a YouGov poll saying that almost half of Britains hate them -> https://www.theguardian.com/commentisfree/2013/aug/01/in-def...
followed by a piece in The Telegraph about how that blog ruined the ultimate upper-class fashion statement -> https://www.telegraph.co.uk/men/style/look-at-my-fing-red-tr...
I wouldn't expect someone from "new" money to wear red chinos.
In stratified cultures, it means "rich-looking person is important", but also can convey more meaning, as the small variances that are allowed take on meaning.
In less stratified cultures, you get jokes about the sales schmuck who lost a sale because he assumed the man was buying the car.
They seem to think they will be working class for the rest of their lives and there's noticeable hate/resent for the rich.
Being an immigrant, I didn't notice any of this stuff. I didn't notice the dismissive polite attitude, I thought it was just politeness. You know, when someone wealthier talks to you, they're polite but there's subtle signals that they consider you "inferior".
I only saw the opportunities, of which there are plenty. I didn't see classes, I just saw that it is possible to go from working class to upper class, much easier than in most other EU countries.
I did start noticing these "subtle signals" after a year and a half or so, but honestly I just decided to ignore them. Since I would always be a foreigner, people would just cut me some slack for not understanding stuff (this also happens in Japan afaik).
Clothes play an important role, surprisingly (OK, maybe not that surprising). If you're in dirty workwear and a hi-vis vest, you're pretty much considered bottom of the barrel. With the right clothes, you can turn into anything from a "chav" or a middle class manager to a millionaire (though you also need the attitude - no slouching, make eye contact, etc).
Just some of my thoughts.
Your observation that the subtle cues seem to be the only thing keeping the classes apart is precisely the goal. This way it costs minimal effort to keep the stratification in tact.
Told to me by an Irish person:
You know the difference between an Irishman and an American? The American looks up at that big, shining mansion on the hill, and says, "someday... someday... I'm gonna be that guy". The Irishman looks up at that big, shining mansion on the hill, and says, "someday... someday... I'm gonna get that bastard".
As an American living in Ireland 8 years now....
Yup.
Disclaimer: my knowledge of British culture is horribly outdated, and mostly draws from George Mikes' "How to be an alien" and TV shows.
I always thought that truly high society in Britain sometimes wears rags, and the neat clothing was a sign of either working class or "new riches". George Mikes makes fun of this:
> On Sundays on the Continent even the poorest person puts on his best suit, tries to look respectable, and at the same time the life of the country becomes gay and cheerful; in England even the richest peer or motor-manufacturer dresses in some peculiar rags, does not shave, and the country becomes dull and dreary.
But it's not just his opinion; for example in the TV show "Traitors", set in postwar London, a character coming from a high society family derides someone for dressing "too well", signalling lower class/new riches, and because of how he pronounces "opera" (according to her, it's "opera" for the lower classes, and "op'ra" for the higher classes).
The subtle signals are about who your parents are, which of the different educational streams you went through, and - more than anything - who you know. Clothes will display all of that indirectly, but the core is always about families and networks.
The idea that opportunities exist is very much a middle class mindset, and anyone who pursues it will inevitably run into the class ceiling. You can make a huge pile of money but it won't buy you entry into the upper classes. At best you'll be a nouveau "someone we can do business with" - or the technical help.
Under the considerable surface polish and politeness the defining characteristics of the upper classes are effortless social - not just professional - confidence, personal entitlement, and exceptionalism. Members are strongly encouraged to believe in all of the above from birth, and it's very hard for outsiders to understand this. There's also a curious affinity with more successful - let's say almost managerial - working class criminals.
I used to live in an upper class enclave, and it was quite astonishing how much time these people spent trying to screw each other over with disputes over family inheritances, property and land rights, various investment scams, and so on.
"Class", by Paul Fussell, is amusing, if dated.
You can't really do that for "traditional" definitions of upper class (which isn't based on anything as simple as having loads of money).
And anyway - everyone is middle class these days. Nobody claims to be working class and there are so few actually upper class people that most people will never meet one.
I think "Working class" is a distinct thing from "Lower class."
At least here in the US, I know a fair number of people that would claim the label "Working class", though I doubt any of them would claim to be "Lower class."
"Working class" is a synonym for "Blue collar" to many people.
"Middle class" is less of a synonym for "White collar", but I think that distinction exists for some people.
It's fascinating to see the differences. One side has flash cars, the other ordinary hatchbacks. One side is happy to wander around in a dressing gown during the day whilst the other is always smartly dressed. One side let their kids play in the street whilst the other has them in studying. One side is mostly trades people whilst the other works in offices.
Despite all the differences (or perhaps because of) it seems to work. I doubt either side would chose the situation but the fact the system forces them together is genius.
So to answer your point, no, everybody is not middle class.
https://scontent.fqls1-1.fna.fbcdn.net/v/t1.0-9/88175441_110...
they'll continue to make a career in entertainment, never transitioning to posing for banknotes like the woman in this picture:
https://dynaimage.cdn.cnn.com/cnn/q_auto,w_1100,c_fill,g_aut...
(Both backgrounds have automobiles and feature greenery, and the men in them all wear trousers. Are there any other similarities?)
Brave New World's "I'm glad to be a β" is funnier if one takes it as an observation, like Cleese's and the 2 Ronnies' 60's treatment, of the actual british class system as it existed in 1930. Having read the M-M-Mitford sisters as well as White Mischief, "everyone belongs to everyone else" may not have been ethnologically that far off of the faster upper class sets. Compare note 3,5 of https://news.ycombinator.com/item?id=24250082 . On a grimmer note, an irish friend told me that the wide eyes of fetal alcohol syndrome were obvious in many faces of the generations before hers.
Is there a name for the "straight man" who motivates an infodump in fiction? In the Yes, Minister series, Sir Humphrey Appleby GCB KBE MVO (as he is now, Winchester) frequently has an excuse to infodump because Jim Hacker (as he was then, school unknown) went to the LSE and therefore couldn't possibly be expected to know how things are done. Fortunately for Mr. Hacker, Sir Humphrey platonically helps him to recall the ideals of government, e.g. "When there is a Labour government, the education authorities tell them that comprehensives abolish the class system and when there's a Tory government we tell them that it's the cheapest way of providing mass education; to Labour we explain that selective education is divisive and to the Tories we explain that it is expensive."
When Japanese people are purchasing foreign products at least, they are responding to the same pressures and cultural biases that guide their purchase of Japanese products and services. In my experience this is stuff like age and legacy of the company, how much other Japanese people like and trust the brand, and all the other arbitrary things that shape the ebb and flow of product and service popularity in a given culture.
This is not to say that there aren't unique aspects to how Japanese people choose products which maybe overlap with the English, but in and of itself the markers of status in Japan are distinct from those in the U.S. (for example), so it doesn't seem likely that Japanese people somehow have some sophisticated notion of not just how to find products that have high status but high status according to U.S. consumers. It's silly simply taken at face value.
I do think the question of whether the English and Japanese share similarities because they are island nations with a royal family (and more) is interesting, though.
EDIT: I should be clear that I'm responding to the entire section of the piece, not just what you highlighted:
It's something like: if you want to know what's good about what your country makes look at what the Japanese and British import. (I can't find the actual quote). And his reasoning is something like: Japanese and British cultures are so hierarchical and stratified, in such long-lasting and subtle ways, that they've become incredibly good at discerning status from tiny signals.
Also worth stating that "what's good" != high-status, so the whole thing seems confused to me.
There's then "Standard" English, where one is expected to utilise the language correctly.
Beyond that, you get into demonstrating a classical education, with the inference that knowledge of Latin phrases ipso facto means intelligence.
I'm much more likely to trust a source written in Standard English as you'd find in a newspaper, as opposed to a poorly written FB post.
I’ve been reading some classic English books, and it is really noteworthy that the writers use Greek phrases and references to Greek plays, clearly as a sign of status (synonym: culture). The οἱ πολλοί are not educated.
Having one’s English corrected is reviled by many, because the implication is that they are poorly educated and lower status. Take care when correcting anyone’s English online or offline!
Another anecdote: In the 1950s the KGB thoroughly infiltrated the British government and security services, but their assets would send back would be public school English so convoluted it literally could not be read by their handlers at times.
It wasn't until the 70s that there was a proper effort to weed out the communists.
I would bet £10 that Hollis was a mole.
This is a pretty good (if not perfect) allegory for working classes / middle classes and upper classes. Those with power want to hold on to it - so entry to power is gatewayed through things like birth obviously but also a series of educational and life choices that prove you are not here to burn it all down.
Social hierarchies exist to both keep the powerful in power, but also as a contract against anarchy. After Brutus murdered caesar he expected to be greeted as a saviour of the republic by the populace of Rome. Instead they all shut their doors and withdrew from the streets for days as they knew what was coming - the chaos and destruction of anarchy and power fighting for a vacuum.
But we should still have our rage. even in the face of anarchy. I think.
Revolutions are the worst possible move. They cause more harm than good and generally involve orgies of injustice and a swapping of a crusty old guard with an even worse power-hungry band of criminals and Jacobin savages. Typically, revolutions draw their monstrous energy from the envy of the haves rather than any kind of injustice. Even when inspired by a desire for justice, it come from a desire to immanentize the eschaton. This always ends badly.
Sometimes, it is better to suffer the shabbiness of a predicament. Building healthy communities and a healthy culture to suffocate the corrupt order with goodness is more constructive and ultimately stronger.
If the SERPs are filled with garbage, their search algorithm loses competitive edge. We may need to go back to a reputation graph-based system. That sort of thing doesn't scale, and Google isn't doing this at all.
In fact, all of the platforms may suffer. How will Facebook keep out GPT-3 spam?
I mean, for a while Facebook may get paid for presenting ads to GPT-3 spam bots, but eventually advertisers will catch on that the people are gone.
Maybe we can unleash bots that appear exactly as humans, ruining the advertiser gravy train forever.
If our weapons (programs, scripts, "AI") become half as effective as theirs (the advertising giants'), they'll drop nine-plus figures paying to have ours outlawed, guaranteed. The only thing that might save us from that in the US is that national ID of any sort is pretty unpopular across party lines, and that's effectively what it'd be.
I wouldn't be surprised if they're already quietly laying the groundwork for such an effort.
In the US it's not possible to catch a flight across the country or obtain a loan without providing your SSN. And the SSN was originally, extra-explicitly, not allowed to be used for anything but administering social security benefits.
Despite what might be described as an absolutely huge market signal that there's desire and need for this, resistance to actual national ID is so high among elected officials that propositions for it have, so far, always been DOA. Maybe national IDs really are terrible and it's worth the shared-over-the-population pain and expense not having one causes in a modern society and economy, IDK.
I'd guess it's one of those things where network effects make it hard to get off the ground, and given the nature of it, especially hard to get off the ground without government buy-in from the beginning, as so many of the ways it'd be useful for saving time and money and reducing risk are tied to interaction with the government. If a market-based solution for it were desirable, for whatever reason, defining some clear and interoperable standard at the government level would probably be necessary to make the market viable, no matter how much latent demand there may be.
Further, imagining the user experience of a non-heavily-regulated/standardized ID market, it seems like it'd necessarily be so hellishly bad it'd never gain adoption. Having 5 different IDs and sometimes having to get new ones or drop an old one seems no better than using credit cards and our social security numbers as IDs, like we do now. It seems to me the epitome of something that can't possibly be any good as a market solution, unless it's so monopolized or regulated that it's indistinguishable from a government solution, except somewhat less straightforward.
The idea that "You won't believe what <popular/unpopular> celebrity said!" or "Trump outrage #942134!" gets tons of engagement when written by a human, but will make people leave if it's generated by an algorithm is hard for me to square with the social media landscape that actually exists.
If ad impressions & clicks don't drop, Google, FB, and others won't give a shit where the content's coming from. If impressions & clicks go up (not an unlikely outcome, I'd say) they'll even encourage it.
That's already the case. GPT-3 might be faster, but there's literally millions of humans producing absolute garbage for pennies. I'm talking 1000 word, quite readable articles (that make little sense if you pay close attention) for $5 or less.
Used to be a time when it was really easy to rank at the top with just that kind of content.
But Google managed to filter all that out, for the most part. Pretty sure they can deal with computer generated text, too.
Hyperlinks still play an important role - if someone links to the article's URL, that automatically increases its rank. No one links to trash. Links from known spam websites will downrank the article or the whole website really fast.
Timestamps used to be abused back when Google preferred the newest stuff - just update it every day or hour and it's the freshest article in the world in Googlebot's eyes.
It's a really impressive piece of technology, and they have a ton of experience fighting abuse on their SERPs - it's happening constantly and it will continue to as long as search engines exist.
These models could be trained on any arbitrary corpus of writing, from any one author or combinations of authors.
Let's see the the critics complain about "boring flawlessness" when the model is trained on the works of Shakespeare, Voltaire, Buckowski, Lewis Carroll, Mark Twain, or Hunter S Thompson... or some combination of the above.
So these authors were drowned out in the noise.
I'm talking about training on just a relatively small number of authors, so the model can acquire some of their distinctive traits.
It's very possible for human authors to imitate style. In fact, if you write a lot of the same kind of thing that you read, you'll find your own style gravitating toward that of whoever you're reading at the time. I just (re)read Franny and Zooey, for instance, and for days I've been fighting Salinger outbreaks in my own prose. Presumably a machine could do the same thing, but I'd guess it's a very hard problem on top of creating something like GPT-3.
edit: and this is not even getting into the thematic and semantic richness of the content, which is something good authors pack a lot more of into each word than the blogspam wordsmiths GPT-3 is about to put out of work.
Usefulness is in the eye of the beholder.
A model trained on, say, all the works of Shakespeare might not create something that can pass for Shakespeare's lost work, but might be inspiring, interesting, or have a Shakespearean quality about it in small doses... which could still be very useful for a writer seeking inspiration.
Incidentally, all of Shakespeare's works make up a not insignificant amount of data, though obviously it's a lot less than what GPT-3 was trained on.
Other prolific authors might also provide enough data to train a model.
It's not clear to me at what point the amount of data one uses becomes too little for useful results, and it would surprise me if there was any consensus on this matter, considering the enormous variety in AI techniques, combination of techniques, and parameters that can be used for training and testing, along with all the different uses one might put the models to after they're trained.
I think the context here is the question of what "kinds" of writing GPT-3 and similar technologies can and can't do as well as humans, so when I talk about "usefulness" I'm really talking about whether GPT-3 can do its thing of generating passable text in response to widely varied prompts. If it can't do that in some domain, then it's not going to replace writers working in that domain.
That doesn't mean the output won't be interesting, but it will be far less relevant in the ways that make GPT-3 notable in the first place.
Style is just another aspect of text it learns to model and predict. If there are style cues, it makes that part of the prediction and continues it; if there are no style cues, then it just starts somewhere, perhaps with some style common on average...
But if there's a decent amount of text clearly linked to a specific author or style, it will learn the quirks, the vocab, the topics, and so on. Shakespare, no problem, of course it can do Shakespeare pastiche - it can do Paul Graham pastiche, it can even do gwern pastiche! Read through the samples in https://www.gwern.net/GPT-3 and tell me again that everything is just "drowned out in the noise" and GPT-3 is unable to produce anything but a bland average style...
I'm not at all convinced. A car brochure at least has to tell you what's good about the car. Poetry seems to actively aspire to saying nothing substantive, and so does a lot of "literary" fiction and a certain kind of magazine article.
Modern marketing is all about emotion.
`S1, E20: The Big Goodbye: The crew of the Enterprise-D sets aside all of its other duties to try and solve a murder mystery. Filmed on location in Miami. A woman dressed in a Starfleet uniform from the ship is murdered, and Riker's going to need a tan to solve this case.`
GPT-3 will reduce the premium on style compared to substance.
You can make GPT-3 affect any writing style. But you can't make it actually have something to say. That feeling - of reading something that's wasting my time, because it doesn't actually have anything to say - is not going to go away just because what I'm reading says "dang" every so often.
These articles may have substance, but simply be far from the best presentation of that substance. I think GPT-3 and similar projects will do a fine job at subpar regurgitation of existing info that's better covered elsewhere but still manages to capture eyeballs, which describes, I expect, something like 99% of all writing on the Web, including, and perhaps especially, message board posts like this, as people often remark when yet another 300-post thread hits the front page on [some tired topic the discussion of which plays out the same every time]. Long-form print isn't perfect but is somewhat better, since there are some barriers both to publishing and to reading and you're not giving the writing away for free so simply holding a passing reader's interest for two minutes means nothing. The web, though, and maybe even magazines? I wouldn't bet against machines doing much or even most of that writing within a decade. Consider: how much of an issue of, say, Cosmo consists of light re-workings of earlier, recurring articles? People already joke about that kind of thing. Machines can probably do that work, very soon.
GPT will serve as intellectual humiliation. Some people will be embarrassed to find out that most of their reading materials can be generated by robots. On the margins, that can lead to people deliberately seeking out more intellectual content. That includes long format materials such as books. On the labor side, writing talent will be allocated away from shallow topics. That's a plus to me.
Seems like all GPT can produce is an infinite supply of shaggy dog stories. I don't know if I'd call that intellectual humiliation, just a lack of a point or punchline.
If GPT produces better output and is carefully mixed in with existing stories, I don't think anyone will notice.
What percentage of stock price stories are generated by a bot? Off the cuff and totally guessing, I'd say 50%. Do you know the exact number? Could you tell? Would the headline service tell you? (probably not)
As for the stories that merely describe stock movements without any real analysis, I myself would read those. I'm completely fine with that as long as there's reliable QA on the output.