#in some ways i think the problem is less 'inconsistent sizing' and more 'inconsistent sizing that's sold as consistent'

[distractionactivated]

As a long-time shopper in both aisles, unfortunately the grass is not very green over in the men's either lmao.

Your waist is 30 inches? Okay, well, our 30 inches aren't your tape measure's 30 inches. Also, we've perfected the technique of making trousers that are too tight for movement but that also fall off. You're welcome. Also, do you have legs shorter or longer than our standard inseam length for that waist? Fuck you, trip over your own trousers/enjoy your cold ankles.

Or take shirts. Your chest is 40 inches so you're a 40, right? Well, yes, sort of. But your shoulders are now too broad or too narrow, you're fat, and you've got too thick/thin a neck so ties won't sit right. Some shirts are sized by neck! So you're a 16.5 neck size. Now it's falling off you because you've got a broad neck on a slight frame. Go down a size and you're choking.

The answer is, just like with women's clothing, to find the one shop that has a pattern block you fit and hope they don't change it. Which they probably will.

Sure, there are usually okay pockets. But that's the only bit of proper, unalloyed green. Solidarity across the clothing aisles, folks.

[would-this-pokemon-be-a-friend]

you know,with the rotom dexes and phones im wondering how good of a pet they are

That is a good question, and a bit of a tricky one. See, there seems to be some inconsistency between the behavior of wild rotoms as described in the pokédex and the rotoms we see working alongside humans all over the Pokémon world. But we’ll get to that in a moment.

To begin with, rotoms are a great size for a pet: They’re both tiny and not confined to the ground, meaning that space will be no issue. That being said, I would advise not touching your rotom when they are not possessing a tangible electronic, as they are known to do, as their bodies are composed of plasma (Diamond/Pearl)! Move-wise, rotoms could potentially pose a bit of a problem. Like a lot of electric-type pokémon, they have the capability to paralyze targets with electrified attacks like Discharge and Thunder Wave. Thankfully, it seems like rotoms aren’t particularly violent by nature, so the usage of these moves on humans would most likely not be malicious. That doesn’t mean, however, that they wouldn’t use them. Rotoms are described in the Pokédex as being avid pranksters (Platinum). This is a pokémon that loves to cause mayhem: I think it’s not too far out of left field to say that a rotom might delight in temporarily paralyzing you for laughs, not being aware of any potential health risk that could pose.

Rotoms posses the unique ability to enter into some machines to take control of them, gaining a new typing. Rotoms have been described as using these new forms to explore new avenues in pranking, some of which could be really dangerous or just plain annoying. Rotoms that posses an oven, for example, like to “sear the surrounding area” (Violet - Heat Rotom) and “gleefully” burn clothing (Shield - Heat Rotom). Rotoms that posses a refrigerator might freeze your bath when you least expect it (Shield - Frost Rotom), while those that posses lawnmowers will mow down every flower they see without hesitation (Shield - Mow Rotom). Essentially, if the pokédex is to be believed, anyone caring for a rotom is going to need a lot of patience and good insurance.

As you alluded to in your request, however, there is a bit of a contradiction between observed rotom behavior and that described in the pokédex. In regions all around the world, thanks to a recent invention by an unknown boy (Sword), devices are utilized that make use of rotom’s unique ability to posses them. Traveling the world you’ll find rotoms in pokédexes, phones, drones, and even special PCs called Rotomis. These rotoms seem to be less inclined to making their owners’ lives a mess, less prone to pranking. This brings up an important question: is the pranking behavior described simply a cry for attention? If cared for and loved by a human, does a rotom become more friendly and co-operative? Or is it that only the most reserved individual rotoms choose to fill these roles? We may never know.

Overall, caring for a rotom has a pretty decent chance of being miserable. There doesn’t seem to be an easy way to keep a rotom from possessing whatever machine they’d like, and their pranking efforts are unrestrained and occasionally cruel. I like rotoms, don’t get me wrong, but I personally don’t want a pet that will hop into my computer and delete all my files before doxxing me on the internet and blasting music all day just for fun.

[lara-prism-light]

Ninjago Rewrite - World Building

For worldbuilding, thanks to new information provided by Dragons Rising, we have a better idea of ​​how the other realms came to be, but in this case it's important to focus on Ninjago in the moment.

The worldbuilding is a problem because it's as if every supposedly important new place in Ninjago was only introduced in the season in which they are important. For example, ninjas have no way of not knowing about Shintaro since, Apparently...

Shintaro is located close to Sticks City and The Waling Alps! Both locations introduced in Possession!

[This map could very well be wrong, so don't take it too seriously, this was just the easiest map to analyze]

There's nothing wrong with new locations being introduced in different seasons, but it's a problem because it ends up being inconsistent.

And I know world-building wasn't on the show's writers' minds as it was initially only supposed to be two seasons, but that will be taken into consideration as I'm rewriting Ninjago as if the lore and world-building were already planned.

A more concrete way to introduce these elements that would be important later is to actually give a basis that this universe is broader than we know. Like, we don't know much about the culture of these different kingdoms, just the little that was shown to us and for me it wasn't quite good enough.

Just a mention of other realms like Shintaro and the jade palace might work, speaking of the jade palace...

Man, how I hate the poorly explained inclusion of the palace and the royal family! Like, just including them in the story as background and world-building information would be enough for the inclusion of the palace to be realistic!

We will do it this way, the palace has existed there since the founding of Ninjago, the royal family are the founders who initiated the city and that is why they are so important. The palace is not in the center but very close to the people, they are the ones who guarantee unity in the city, the mayor is the one who maintains order and the Borg family is the one who provides technology over the years. Thus creating a sense of importance for the three groups, thus making their roles as important figures in Ninjago City that much more important, Because ther ancestors were important to the founding of Ninjago City. Both in politics and technological advancement.

Another form of worldbuilding would be an engagement of social statuses for the inhabitants of Ninjago. We only see people from the lower states in Sons of Garmadon and Master of the Mountain, but other than that we haven't seen much of the rest of Ninjago.

Maybe Sticks is a poorer city because it has no connections to Ninjago City and therefore doesn't have the same resources, which would make Ronin steal to bring some of there to his beloved city.

What about the serpentines? We know that they created a society in the sewers of Ninjago City, but even after having helped defeat villains several times, they still live in the sewers?? Why didn't they try to fight for their rights to live with humans on the surface? Honestly this would be a very interesting story line to explore in the show, it could be a message about racism and discrimination, I mean, their story is already about discrimination!

They were discriminated against and treated like monsters for being different, it's no wonder they were ok with the serpentine war!

One more thing is that the island of darkness is very generic and even though it is the same size as Ninjago, it is very empty in comparison. Why after the events of the first seasons don't they try to explore more of the place? Why not become a meeting point for villains? Wouldn't it be interesting to have the villains making a base there already? Since no one goes to that place, much less the ninjas who seem to have simply forgotten about the place? I just think the island needs some use other than decoration at this point.

Why are most important places only known to ninjas or other important characters? Why not other people? Why not Cluth Powers? He's an explorer, isn't he??(and yes I will be rewriting him completely because I think it's ridiculous that they took an iconic Lego character and just made him egotistical asshole!!)

Well, in conclusion I plan to reimagine the world of Ninjago as a whole, kingdoms, cities, culture, important locations, realms etc.

That's it for now, the next part will be about elemental powers and elemental masters.

[oddball-artz]

YOU MENTIONED AN EPITHET!

*cool epic drumroll*

WHAT IS HER EPITHET? SOUNDS COOL, AND DOES IT HAVE ANY MAJOR DRAWBACKS?

(Apologies in advance for how long it takes for me to answer these, creative block is a bitch, tysm for asking abt her though, I can't even express how excited I am to have someone who isn't me or my partner like an oc of mine) Her epithet is soothe, but I'm still working out the details of her major powerset, but I can give you some of her powers and the drawbacks her epithet has (these are subject to tweaks and changes however, and if you have any suggestions I'd love to hear em) :]

She has the ability to heal and soothe the pain from wounds if she can touch it (think the ability lay on hands from d&d). Visually, when she does this, a lavender colored aura surrounds her hands. When someone touches the aura, it emits a sensation close to that of menthol. Drawback to this is that her hands will sometimes go numb with the same menthol-like numbness, but this time hiked up to an uncomfortable degree(kinda like when your arm falls asleep but on steroids). This doesn't happen often, but the more she uses her epithet, the more likely the chance of it happening. It can last from anywhere between a few seconds and a few minutes, and she just kinda has to wait it out. Heat helps the numbness fade faster, so she carries around those snap to heat packs in her pockets just in case.

She has another aura, but this one calms rather than heals. The aura itself is less opaque and more shimmery than its healing counterpart. Still affects the likelihood of the numbness, tho. (This aura is less like lay on hands and more like sanctuary, which is also a dnd ability or Molly's silence bubble). She can shape it and adjust how far it reaches (within limits of about 10 feet). Within the aura, it's been described as feeling like the weight of the world has been taken off your shoulders, even just for a moment. Dalia wouldn't know though, all her powers only work on others.

She also has exactly one summon. It's glowing moth things (that may or may not be inspired by the visuals for Here Comes A Thought from Steven Universe) she can summon a maximum 10 at a time, and they're pretty average size for a moth. When they make contact with someone who's not Dalia, they poof into the same calming aura that she can make, just on a way smaller scale. There's two problems that come with them, though. One, she can't unsummon them once they're summoned, she has to wait for them to poof on their own. And two, sometimes when she's nervous or stressed, they just kinda appear. She's not consciously summoning them, but they are her newest power, so she has trouble controlling them and making sure she doesn't summon them on accident.

I'll probably go more in-depth about it later, but that's the basic run-down of it. (Also, sorry for any typos or logical inconsistencies, I blame it on the fact that a lot of the writing for this was done at ungodly hours of the night lmao)

[tigger8900]

The Black God's Drums, by P. Djèlí Clark

⭐⭐⭐⭐

When Creeper, a young girl scraping a living on the streets of New Orleans, discovers a secret plot to deliver a powerful magical artifact into the hands of Confederate soldiers, she believes she's found her ticket off the streets and into the skies. With the aid of a powerful orisha - a spirit of storms who speaks in her mind and channels powerful magic through her body - and the captain of a pirate airship, can she stop the trade before it's too late?

This bite-size adventure was a fascinating window into an alternate history. As I was reading the first few scenes, historical inconsistencies - wasn't the Civil War over? what were these factions? - began piling up, and I forgive the author for the resulting expository info-dump because it was all so interesting. He clearly spent a great deal of time thinking through the necessary what-ifs and so-thens to create his alternate New Orleans. I especially loved the way the orishas were woven into the setting and plot, not just creating a good story but also driving me to wikipedia after I'd finished the book.  Something else I think he did well was the incidental diversity. It was never particularly important to the story that a character was bisexual or had a prosthetic; rather, they just existed in the world, as they have every right to. The author's use of dialect, which is present on nearly every line of dialogue, might bother some readers. I never had a problem following the speech, but if it's a problem for you, there's no escape from it. One thing I didn't like was less the fault of the book and more a problem with the marketing/cover copy, and that was that I had been promised an airship adventure. Unfortunately, that's not what this book is about. This book is the prequel to the airship adventure(sadly, not yet written), but is itself very firmly grounded. It features sky pirates, but is not a sky pirate adventure. Maybe one day he'll revisit this setting and we'll get to see Jacqueline all grown up and ready to take to the skies.

[ledburywonder]

For the first time...

For the first time since commencing the apprenticeship, I felt the real benefit of a Continuous Professional Development (CPD) opportunity in Conflict Management undertaken recently.  The course detailed at https://www.irpmlearninghub.org.uk/course/view.php?id=54 was the only formal training I have undertaken, despite having had a career where I have faced down guns and knives.  Not, I hasten to add, in a retirement complex.

One of the nearby developments, Bankhouse Court, has had a poor record of retaining staff; there have been four Development Managers (DM) in the last five years.  I have come to realise that in any development, no matter what the size, you tend to see certain character traits appear; the Mother Hen, The Queen Bee, the Serial Complainer, and the Manager are but a few.   In training a DM, I have yet to find a textbook guide on how to deal with them.

With the Area Manager away, I was asked to look at three letters written recently by Mr & Mrs Collins.  It is easy to read a letter as someone registering a complaint.  However, I try to read looking for an insight into the writer’s expectations. By definition, customer expectations are any set of behaviours or actions that individuals anticipate when interacting with a company. Historically, customers have expected basics like quality service and fair pricing — but modern customers have much higher expectations, such as proactive service, personalised interactions, and connected experiences across all channels[1].  Clearly, from the content of their letters they had expectations that were not being met.  They had asked for a meeting to discuss their issues.  I arranged a date and time when I could be on-site and, in a communication to the residents invited anybody else who wished to attend.  My preparations included thinking about each of their issues and reviewing if they could be met under the terms of the Development lease, the roles and responsibilities of a DM, company policies and even the law such as the H&S at work act.  It was apparent that over recent years, some practices had become common which were questionable and misunderstood from what could be achieved.  You break expectations by changing what someone’s already used to. You change expectations by giving them something new[2].

The recent CPD introduced me to the Thomas-Kilmann model of conflict management. This model is used for resolving conflict since it involves considering the most important factors before deciding how to handle the problem. The key is to be more conscious, and less reactionary in the way you handle conflict. Recognising which of the five approaches to conflict management you are most comfortable with and broaden the ways you manage conflict by using the Thomas-Kilmann model. When dealing with conflict, analyse the intrapersonal, interpersonal, and intergroup conflict present in your role, and how it affects your personal and company performance.

Meeting Agenda item:

Concern over the DM’s roles and responsibilities.

Discussion:

Concern was raised by a number of residents that the frequent change in DM’s was resulting in a confused, patchy and inconsistent approach to the building’s management.  The DM may find, in her daily activities, that previous roles are influencing her decisions.  The residents may have expectations that are not being met.

Conclusions:

The importance of clear, concise and friendly communication was agreed by all.  The AM has to ensure there is a consistent provision of service.  The DM is still undergoing her initial training and is daily receiving support from the Area Trainer. When residents have a concern, they need to raise it directly with the DM and, if within a week, they still feel their concerns have not been addressed they should contact the AM.  The DM should be available daily in the site office to hear and exchange views.  The Residents’ Afternoon tea sessions are also an ideal time to listen to what is concerning residents.

Action items

Continue to provide training and guidance - Area Training Manager

An understanding that issues can usually be quickly addressed when you listen, reflect and inquire - All parties

Recognize that other viewpoints are possible and likely - All parties

Be open-minded to creative options - All parties

[1] https://www.salesforce.com/resources/articles/customer-expectations/

[2] https://signalvnoise.com/posts/3035-you-break-expectations-by-changing-what-someone

[idontwanttospoiltheparty]

https://variety.com/2022/music/opinion/lennon-mccartney-taylor-swift-bon-iver-billie-eilish-finneas-songwriting-ai-analysis-1235311633/

Thought you'd find this article interesting 😊

Hey there, thanks for sharing and sorry for taking a while to get to this!

Here's a clickable link for anyone interested:

So, I was recommended this article a few days ago, and it did peak my interest, but I honestly think their findings are kind of lackluster, and at the same time their sample size seems way too small to make much of a general statement. Plus, objective song analysis is already difficult for humans, so I don't fully trust an AI to be able to notice anything particularly interesting (I also happen to be a STEM girl who thinks machine learning is overrated lol – not useless, but also not some be-all, end-all solution to all our problems).

A few things that pinged as weird to me while reading:

OH?????????? In My Life is definitely co-written now, is it? (Laughing, remembering that OTHER AI that claims it's even less co-written than even John said it was – apparently his and Paul's words can't be trusted as much as whatever patterns their AI picked up on, even though these patterns were necessarily based on songs that we know were written by Paul and John based on THEIR WORD. Not to mention that that AI doesn't seem to account for collaboration, where say someone finds the chords and someone else finds the melody, but I digress)

Honestly, you gotta show me which Paul Solo songs you consider NOT sexy before I believe you on this lmao. Does the AI recognize innuendo? Like, maybe it's true, but 🤨🤨🤨

The comment about the time to reach the chorus plus slang use is interesting to me, but like what does "lyrical density" mean? Less repetition? More words per second? It's so vague!!!!!!

Moving on…

Okay, why are we analysing a songwriting duo with only two co-written songs? (Also, William Bowery AKA Joe Alwyn is co-credited on the two collabs they're counting for this, evermore and exile, but they're not counting the song Birch, which was written by Taylor, Justin Vernon/Bon Iver, as well as Bryan Devendorf. There's some inconsistency going on here.)

Like, the sample size here is ridiculous tbh? Like, what's the point? Why not Taylor and Jack Antonoff or Taylor and Liz Rose or Taylor and Aaron Dessner, where there would be at least about a dozen songs to pick from? Still not ideal but a lot better.

Yeah, except the two songs are off Taylor albums that both notably and frequently use uncommon vocabulary? It isn't unique to this one duo/it's a reductive statement that ignores context.

On to the Billie Eilish & Finneas part

Actually, back up, what are these categories? How does this AI differentiate romance, love, and infatuation?? What's the point in this when you're speaking in the vaguest of concepts?

In conclusion (?):

UNDERSTATEMENT OF THE CENTURY!

Then it closes with a vague statement about the current breakdown of genre walls, which has little to do with what this article was actually about so All Righty Then.

Overall assessment: yeah, I'm not very impressed at both the lack of musical and scientific nuance here, it's a bit of a lose-lose for me, but thanks for bringing it up I had considered making a post about it but figured no one would care lol.

[thecrackedbead]

Catching Pearls and Iron - Honouring the old by making something new?

---

Left: The original. Right: The final.

---

Hmmm... That looks like it was done by hand, I think as I spot a bright red bracelet and my hand bravely ventures into the tangled vines that make up the thrift store's jewelry section.

The thrift store is hardly a rare location for me. I take pride in being a thrift junky. I love finding treasures among the ruins. Sometimes, the things found shine as bright as diamonds. Other times, they look like they belong on a midden heap. My initial thought is that I found the former but, when I withdraw the item, I am not sure that it isn't one of the latter.

The bracelet's backing is made using of seed beads of inconsistent size (supposedly the same size but definitely not). Onto this, a variety of nuggets are attached. Some of the nuggets seem to be proper stone. Other red ones are some sort of resin and scratches have marred them and dirt is burrowing into them. Even if this was not a problem, the over all colour combination is 'interesting'.

Is it ugly? Is it beautiful?

It is not to my taste. However, taste varies and some careful tugs on the bracelet show it to be strong. Good workmanship and, as I have yet to make anything with netting technique, I am intrigued and I am not beyond buying old jewelry, then selling for a greater price.

(In the past, I have had guilty feels about this last thing, but I eventually came to the conclusion that antique brokers and art dealers do it. As long as I make a point of saying I am not the original artist... and I am planning to fix this bracelet up before reselling.)

Inevitably, I shell out the $5.99 for it and take it home.

One week later, I fish it out from my Hoard of Doom (TM).

Hmmm... I think. First, things first: remove the scratched-up red resin beads.

Instantly, things don't go to plan. The red beads do come off, but so does a number of the other protruding beads do as well as the entire beaded toggle clasp (and *sigh* naturally get sprayed all over my floor). They are all on the same thread with apparently little in the way of extra security. I could fix this. I coooould, but I am not the biggest fan of the other beads and feeling extremely slightly guilty, I proceed to remove them all.

---

Left: In case you don't know what a toggle clasp is. Right: I'm not going to crafter's hell for removing those beads, right?

---

It leaves me with the bare netted-back minus the clasp. It doesn't look half bad with just the back and that leaves me with two possible plans:

Make another toggle clasp and call it a day.

Sew more beads on then make a clasp.

Naturally, because I am a masochist, I pick #2, and because I am anal about picking the right beads. I haul out nearly every kind I have and, because I am a clutz, manage to decorate my floor with a goodly many of them along the way.

When they are properly set before, I end up stuck. I know that I want the protruding beads all be in one colour as I felt that the variation of colours in the backing was more than enough. However, I still have a number beads that would work but can't settle on one. In the midst of this, the ghost of guilt due to disassembling someone else's work is casting its pall over me. Surprisingly, this helped me.

---

Tiny fraction of the beads I considered.

---

The original protruding beads were of many different sizes and colours and I have quite a variety of shapes of hematite. The shapes are nowhere near as varied as the original ones, but I like to think that by using them and purposely having them randomly arrayed on the seed bead backing, I will be honouring the original work and so its creator in some small way.

I also decide to add acrylic pearls to the work. Why? Firstly, I like the variation they add to the mix. Secondly, they have been sitting in the hoard for almost a year and need using up. Secondly, hematite, being made from iron is heavy, and by filling up some of the piece's space with light acrylic means that I am less likely to sprain someone's wrist by proxy.

When you finally find what you're looking for... I keep on feeling that there ought to have been stars and birdies floating around those beads.

---

I would like to say that everything went smoothly from here on. That I didn't have my beading thread get so hopelessly tangled with the diamond shapes at the sides of the work that I had to redo vast section hundreds of times. I would like to say the former didn't also happen with the protruding beads and my thread at least as often. I would also like to say that the previous crafter's thread wasn't a thick monofilament that meant that getting my needle and thread through the beads was a huge nuisance. This thing did not cause me break multiple seed beads which didn't mean I had to scrounge matching beads and thus redo EVEN MORE of the work. I would like to say that modifying this bracelet did not take nearly three times as long as making a new one. I would like to say that all the previous did not mean my swearing provided a brisk musical accompaniment to my crafting.

I would like to say that.

So I will.

Everything went smoothly from here. I didn't have my beading thread get so hopelessly tangled with the diamond shapes at the sides of the work so that I had to redo vast section hundreds of times. The former didn't also happen with the protruding beads and my thread at least as often. The previous crafter's thread wasn't a thick monofilament that meant that getting my needle and thread through the beads was a huge nuisance. This thing did not cause me break multiple seed beads which didn't mean I had to scrounge matching beads and thus redo EVEN MORE of the work. Modifying this bracelet did not take nearly three times as long as making a new one. All the previous did not mean my swearing provided a brisk musical accompaniment to my crafting.

One last thing on this subject. There is most definitely not a stick protruding from where my nose should be and it is most certainly not growing as I type this.

---

The bracelet from Hell. An easy bracelet to make.

---

Whatever erroneous rumours you may have heard on the process of making this bracelet. I am pleased with the final result and I'll likely make other bracelets in this style.

If the original crafter sees this, I hope they do not begrudge what I have done and I would like to thank them as I likely never would have made it if not for them.

---

Now for sale at my Etsy store.

[itsclydebitches]

“They dumped thousands of people into a desert and those people were getting picked off by grimm until Winter arrived” - Atlas alone looks like it should have a population of millions, the fact that Mantle is the same size due to being in Atlas crater leaves to believe it has a similar population. Maidens aren’t all powerful, I refuse to believe Winter and 3 people can protect a crowd of MILLIONS. More are most definitely going to be picked off.

Yeah, consistent numbers are another thing that just don't exist in RWBY. On the one hand, I'm pretty sure Cinder told Salem that they managed to evacuate "thousands." On the other hand, you're right that Atlas and Mantle look way bigger than that. Now factor in the that we never see any soldiers evacuating and, according to other fans, there's a moment when Oscar mentions Remnant's whole population being in the "millions". What we're left with is a lot of inconsistencies lol. Are we meant to read this as the Kingdom of Atlas actually only having "thousands," Team RWBY successfully evacuated everyone, and the show is just really bad at animating city sizes and remembering to include soldiers? Or are we meant to read this as the Kingdom of Atlas having "millions," with Remnant some larger version of "millions" (Atlas is, after all, a pretty big portion of the world), and Team RWBY massively failed in the evacuation, resulting in only a very small portion of the population getting to safety? I honestly have no clue. Though I am more inclined towards the second reading. There's no army, Maria and Pietro were forgotten, Qrow's team was forgotten, Cinder's comment makes it sound like they only partially succeeded... the overall picture is one of a less than stellar evacuation.

Regardless, whether it's "thousands" or "millions" ultimately doesn't matter when it comes to their current problem. I agree, I don't think Winter alone can defend that many people. At a certain point the numbers become inconsequential because they're all in the realm of "too many for a single fighter to defend." And it nearly is just a single fighter. Ren's aura is gone. Oscar's trump card was the cane he's now used and he's otherwise done in from being shot, falling, Hound attacked, and tortured. Winter may be the Maiden now, but she just came off of Cinder's attack, two days of non-stop war, and two fights with Ironwood. Emerald is the only one who is maybe still in fighting shape here. If RWBY were to treat this situation at all realistically, it doesn't matter how much Maiden power Winter manages to tap into, or whether Emerald is at her strongest, or if Oscar dips into a latent well of energy, or if Ren is still willing to fight despite any blow potentially killing him. Even if they all overcome the limitations they should be experiencing, the four of them cannot be in thousands of places at once. Even a Maiden can't protect that many people when you've got grimm coming in from all sides and a storm messing with visibility. If they don't find Shade soon, a lot more civilians are dead.

[masterthespianduchovny]

Let’s take age off the table, I know shocker for me, the other thing that bothers me about this development isn’t that it can’t develop into something serious without major sacrifices happening if they stay together.

And that doesn’t it right with me.

Society loves romanticizing that “if it’s true love, a sacrifice doesn’t feel like a sacrifice.”

Girls are especially conditioned with this belief. On average, girls and women give up far more for love and a relationship and a significant amount gross to resent this in some capacity. They have to tell themselves that their relationships and kids made the sacrifice all worth it. And can’t express regret at the road not taken.

It’s why the saying, “you can’t have it all” is mainly directed at women and almost never men. Because, in comparison to women, men tend to have it all. Men are less likely to give up careers or take a step back in the name of love. And, when they do, they typically have more opportunities for them than women.

And the same goes for people of color. Their is rampant racism and hiring discrimination, so Black people have to move more carefully professionally.

Which brings up to the Sam and Rebecca storyline.

Both of them love and are invested in their careers. Their relationship jeopardizes either or one both of their professional goals and desires.

The idea that they can and will get over this because “love” is so dangerous because this relationship had massive red flags and was problematic from the word go and they both knew and know that. It’s not like they dated without personal attachment, and then a professional opportunity presented itself.

No, they voluntarily entered into a relationship that can only end in disaster.

If they stayed together and went public, there is no way in hell Sam could stay on the team.

Which means Sam would be the first one to take a professional hit. Sure he could go to a new team, but we wouldn’t know where that would be, if he would have to be traded/loaned, and if where he went would even be beneficial for him as a player. There is also his rejection taking a hit, because even tho they will pat him on the ass, his professional cred will have taken a massive hit.

And, for real, what owners do you think will be okay with a professional player having that kind of access and intimacy with another owner?

Sam’s professional development would also take a hit. Remember how Sam was struggling with Richmond at first, until Ted problem solved why that was? Remember how his stats was taking a hit? Remember how ted constantly advocated for Sam as a player and utilized him in a way that was appropriate for him and, in turn, his confidence not only returned and flourished under Ted.

And we saw how Jamie reacted when his loan was terminated. We saw the difference of play and what it did for his confidence training under ted v not.

I’m not saying Sam needs ted and can’t succeed without him, however, coaching arent a one size for all. Some coaches work better for some than they do for others. Some players need certain types of coaches and only respond to one specific coach.

Sam’s development is threatened by not only losing a coach, but having to play for a new team, learn a new system, and make bonds with new players. It may not seem like it, but switching teams, esp when don’t want to, is hard on players.

Sam would be voluntarily setting back his career to be with Rebecca, which may harm him in the long run.

With Rebecca, most have noted that she’s been treated like shit in the media and, in this case, it would be rightfully so. We shouldn’t be arguing that just because Rupert’s gets a pass, Rebecca should too. We should be arguing that Rupert needs to be criticized just as much if not more for what he did. Professionally speaking, Rebecca actually deserves far more criticism than Rupert. He just ran a mediocre club, she’s unintentionally undermining it by dating a player, which reeks of massive irresponsibility on her end.

Some have argued that since Sam and Rebecca have great chemistry and this could be something great, maybe Rebecca should sell her club.

Fucking, excuse me?

Regardless of how Rebecca originally felt towards the club, she has grown to love it and has a deep attachment. She even takes genuine pride at her ownership and trying to get them to succeed.

Some are really saying that a woman should sell her club to be with a man she shouldn’t have gotten involved with to begin with? Like, I’m hearing that correctly??? (Yes, this has been said.)

So in a profession where women rarely own clubs and a professional she enjoys, is a because she should give up for love?

And if this relationship doesn’t work out, which it most likely won’t, them making these massive sacrifices won’t be “no big deal, I’m glad I took the risk.”

It would be, “what was I thinking selling my club/fucking over my career?”

Rebecca’s media problem wouldn’t go away with her selling the club, she’d still be dating a footballer and mags would still write about her. This is a fucking juicy story. It would take years before this wouldn’t be in constant rotation. Even if they both went away, this story wouldn’t die and would hang over AFC Richmond like a dark cloud.

Love isn’t going to make Rebecca stop caring about what the media says about her, esp if they’re right in their criticisms this time around. That criticism practically ate her alive. She’s going to feel like the monster she thought she was in season 1.

Especially since this is going to be way worse and more frequent.

And do you imagine that she keeps her relationship with Keeley or Higgins? Do you think she doesn’t become a pariah?

So a woman who already went through a humiliating and public divorce, then gets into a relationship that increases attacks and criticisms on her ten fold, and she sells her club to be with this dude (hypothetically, of course), do y’all think Rebecca doesn’t feel even more isolated and lonely? Do y’all think that Sam’s love is enough to compensate the massive loses she’s taken? That nothing else matters because she has him?

Some want to pretend like it’s no big deal, but either they don’t understand the implications or they think just because it’s fiction it doesn’t matter, but it does. It does matter what they’re saying and what they’re not saying. Because if it’s fiction and doesn’t matter, why complain about how raze is being portrayed as well as gender?

Age can’t be treated as unimportant because it’s a fictional pairing, yet racism and sexism doesn’t. That’s called inconsistency and cognitive dissonance. Either all of these things matter or none of them do and, yes, in this case it is all or nothing.

I’ve explained enough why this age difference matters and to pretend it means nothing because “he’s an adult.” Like man, young adults are truly fucked if older people feel this way. If older people feel their responsibility and obligation for protecting younger people stops at 18.

And, perhaps, this explains why people don’t understand the other complexities and nuances of what this relationship is fucked up. Or why they’re downplaying it.

As someone once said, “all art is political.” This and these discussions are a politic of sorts and so many people are missing the mark, yet are only advocating and having nuances discussions on this topic advocate for a problematic and detrimental relationship.

[todomitoukei]

what's your biggest gripe w bnha? and what kind of plot/character changes would you make if you had that kind of power? (this sounds like a question you get for an interview or an essay im sorry 😭😭 im genuinely curious though because you're very big brain and have interesting thoughts)

Did you hear that, mom?? This person says I’m big brain; suck on that!! Thank you so much, I’m glad you find my thoughts interesting!

Anyway, I would take a completely different approach to the story. Focus less on the heroes and more on the villains (than the story does), and most importantly focus more on the correlation between the hero system and the existence of villains.

Bnha oftentimes feels all over the place in terms of genre and what it wants to focus on. Just compare the first season to the current manga arc. The first season felt more like it was going to be a very light-hearted, slice of life type of story that would just focus on this kid who tries to become a hero. But at some point, it turned into a whole social commentary on injustice and discrimination, as well as touching on other really heavy topics.

While where the story is at right now is much more to my liking than the first season is, it can just drive a lot of people away from it, too, and is partially also the reason why there is such a divide in the fandom - either you like the light-hearted hero aspect of it and then you end up not even paying much attention to the story when it focuses on the villains and you don’t want to see people in the fandom talk about the deeper aspects of the story, or you are more into the heavier stuff and either have to wait a long time until the story gets to that point or just get turned away early on and then never get to the interesting part.

That is not to say that having those two sides is inherently bad; it’s nice when deeper stories also have their light-hearted moments, but it just kind of jumps from one side to the other all the time instead of just focusing on the actual points it tries to make.

I probably would’ve just made the class a whole lot smaller. This story has far too many characters because it doesn’t just focus on the students; it focuses on the students, the Pro Heroes, and the villains, which then leads to the issue of the majority of these characters being absolutely irrelevant most of the time. The class could just be half the size and that would’ve been just as fine and would allow for the remaining students to be fleshed out more and get more screentime than they do.

Personally, I’d prefer Shouto as the protagonist, but Deku could actually be a good, interesting protagonist if Horikoshi wrote him better, which isn’t difficult considering Deku is technically the perfect protagonist for this story. He is someone who has grown up experiencing that injustice of a quirk-based society, the problem is that as soon as he got his quirk, he suddenly seemed to have forgotten all about that. And that is actually fine if it only lasted for a while, but he still hasn’t realized that. He could be a really great character if he just started questioning things earlier. That doesn’t mean he has to give up on heroes and try and start a revolution right away, but maybe have him actually point out how messed up it is that Endeavor is one of the top heroes, when he’s the type of person heroes would usually help arrest.

Fewer side stories! Season four was so annoying in terms of how manga arcs were spread across the episodes. The Gentle arc was far too long, which then led to the Pro Hero arc only partially being in the season; the remaining parts then having to be continued in season five - not to forget that they chose such a boring, anti-climatic cliffhanger?! I kind of understand the purpose of the Gentle arc but it could and should have been way shorter.

Overall, just give more screentime to the villains - the League of Villains, that is. The story could be better executed if Horikoshi actually bothered to focus on the injustice caused by the hero system, but it’s often not obvious enough when the major focus continues to be on the heroes, with the villains only being focused on every once in a while (more and more as the story goes on, but that built up is just really slow). Not to mention how Horikoshi himself treats the heroes vs. how he treats the villains, which incidentally basically parallels how they are treated within the story.

To sum it up, I think if there were fewer background characters and side stories that aren’t relevant to the plot it would allow for a more balanced story by giving more screentime to the villains and their backstories because this constant over-shadowing makes it unclear whether Horikoshi actually wants to have the story deal with these topics or not. 

There are many things in the story that are just inconsistent or straight up contradict something else within the story, so if Horikoshi could write this story a little clearer and get to the point a bit better instead of having this weird back and forth it would be a far better story and one that could be told in fewer pages, considering if we take the current arc some kind of big turning point in the story, then it took roughly 300 chapters to get there. That in itself wouldn’t be too bad if the flaws of the hero system were more consistently focused on before. So better pacing would be nice (and stop throwing every possible reveal at us for no reason); he doesn’t have to drag out how the heroes constantly get away with everything over and over again - we get it! But how are those flaws of the hero system going to change, that’s what I would like to find out, hence the story would benefit from showing those darker sides more.

In short - I would change a lot of things, but the main idea is to cut out the more unnecessary parts and then shift the focus a bit more to have a more balanced story overall that is clearer on the point it’s trying to make.

[solacefruit]

Do you think Warriors would have less problems if it were written by only one author instead of a team? I mean I think there used to be three, now two people who write the books, but they get a rough outline for the books by a group of editors, if I'm not mistaken. With so many people stirring the pot, I feel like it's bound to cause issues. I wonder if Kate and Cherith could write whatever they want instead of what the editors tell them, Warriors might be a bit better

Hello there! I don’t necessarily think a smaller team would solve the core issues of Warriors, but it would probably solve some issues across the series--particularly issues of consistency. 

It’s a common joke in the fandom to point out the fact that there’s about a page of errors for every single book, ranging from incorrect names to different appearances to swapping genders, but to me that level of regular inconsistency speaks to an inattentive editor and a lack of ledger-keeping on the writers’ side. If you’re working with a team, you should really be working from the same master copy of character details--which apparently Erin Hunter does not do. 

So a smaller team might help them keep things more consistent like that, but it’s also worth mentioning that it’s not actually that hard to keep details consistent and reducing group size is a weird way to solve what is clearly an organisation and communication issue. 

But more broadly I don’t think Warriors would necessary have fewer issues if it were written by one author, because that actually depends entirely on who that author is and what is their agenda in writing the series. Plenty of solo authors manage to write terrible books and I personally have little to no faith in the skills or integrity of any of the Erin Hunter collective as artists by this point. I think they are deft and savvy money-makers on a good run, but not artists in the sense I could consider many fiction writers. 

The series is primarily made to be a formulaic and high-turnover, capitalising on a particular demographic’s uncritical enjoyment of cat stories. This is not an “evil” way to make literature, but it isn’t one that I would say is often defined by integrity, creativity, innovation, or a love of art for art’s sake, and I wouldn’t consider the state of Warriors to be a result of “mean editors censoring the authors’ voices” (or something like that) so much as it’s editors working with a writing team to hit deadlines in a rapid pace so as to keep up with demand. 

I’m obviously speaking as someone outside the writing room on this one, so I don’t have any special insight into the series and this is all speculation, but my instinct on this one isn’t authors being thwarted from completing their artistic vision or the result of too many people trying simultaneously to creatively direct the series. My sense is that the books are outlined by editors to be keeping with the successful formula as much as possible and then each writer has a word-count to meet following that outline. 

That probably sounds a little cynical, but frankly it’s good business. It doesn’t make for good stories or good writing most of the time, but when your target demographic are frankly not very good (or at least, not very critical) readers who delight in familiarity, that is rarely an issue. So pretty much what I’m saying is the premise isn’t really “Warriors would be better if written by one person” because I don’t necessarily think that’s true; the more accurate premise is “Warriors would be different if written for its own sake, rather than to capitalise on young readers.” 

Because yeah, if one person--i.e., me--wrote Warriors, it would be extremely different. But it probably also wouldn’t be popular, or maybe even published. The success and popularity Warriors currently enjoys is linked to how it’s manufactured, which means that it requires a team of writers and editors in order to be written in fast enough succession to remain relevant to a rapidly ageing reader-base. 

I don’t know what Kate and Cherith’s dreams for the series are, so I can’t speculate there, but I’m going to say I personally doubt that they’re unhappy with how it’s going so far. Sometimes writing isn’t an act of deep passion and artistic vision: for some people, it’s just a job! 

[godslayingenthusiast]

https://qgpennyworth.com/portfolio/people-are-strange/

People Are Strange is a work from Holy Nonsense, a Creative Commons project. View Holy Nonsense 2020 here.

Each entry (single page or multiple pages of the same work) is released under an individual CC: Attribution, Non Commercial, No Derivatives license. That means you can repost this work as-is anywhere for any non-commercial purposes.

Image descriptions, including transcriptions of text, are expressly allowed, but just make sure you include the credits that are baked into the image when you do them. Image Description after the cut.

People are strange.

I often see people walking around, both in public and on the clock at their jobs, in the strangest costumes.

The guy walking around dressed like a Cowboy straight out of an old Western movie, who has never wrangled a cow or whatever in his life. Who lives in the apartment two doors down from me and doesn’t even know what a pasture smells like. He likes his theme so much he drives the big pickup truck, has a belt buckle the size of Texas (here in Wisconsin), and when he can, he carries a gun on his hip just in case Billy the Kid shows up one day.

The woman dressed like some sort of gothic vampire. With the “I <3 ZOMBIES" keychain. White powdered face with heavy black eyeliner.

The person whose gender I do not know, with their carefully dyed and styled hair and curated collection of buttons and piercings and shirts they stole from a 90s Xena fan’s closet.

The guy in the Businessman costume, not a tuxedo but shirt and tie and "very nice" shoes.

The casual Steve in his old hole-filled ripped up T-Shirt and wholly holey jeans.

The Princess in her stunning dress and accessories and make-up that had to have taken at least two hours to put together each morning.

These are all people I work with. Walking around in costume, covered with the marks of their respective themes. This is fully accepted for the most part. There are exceptions (the redneck likes to poke at the queer one and the casual steve is always ribbing the businessman and whistling at the Princess) but for the most part nobody is ever told to "go home and come back dressed normally". Why? Because that is normal. These are accepted normal ways to dress in modern society, even by supervisors and managers.

Don't misunderstand, I'm not judging or criticizing. Actually I think it's pretty great. But, as fucking ALWAYS, my problem is with INCONSISTENCY. GOD I HATE INCONSISTENCY. My brain latches on to it like a life raft and will NOT let it go.

If I show up to work today wearing a Wizard robe, a wizard hat, carrying a big magicky-looking staff, driving a car painted to look like a dragon, I'm the weird one. Not the Cowboy, not the Princess, me. What makes one theme more acceptable than another? A robe and hat and stick are much easier to set aside, take far less preparation, and are more likely to make people smile, than a full-on Vampire costume. However, if I go to work like that I have absolutely no doubt I would be sent home to change into "regular clothes". And If I went home and put on a cowboy costume and came back, that would be considered changing into normal clothes.

I would like to know who draws these arbitrary lines, and what kind of bribery would be effective on them.

I want to see more wizards, more Monks, more goofy Doctor Who outfits, more people dressed like Neo in the Matrix, driving around my workplace in a forklift. Operating the machine next to mine. Fixing my shear when it breaks down. Instead I get about 30 cowboys and a few of the other mentioned types, and that's it, because anything else is somehow "too weird". Worse, at least half of the fucking cowboys are of the openly racist, kill-animals-for-fun variety. At least the guy in the Wizard robe I know I'd probably get along with.

Or Kill Me.

- Trix

Marginalia: a large, ornate and friendly frame containing the text "Be assured I shall continue doing whatever the hell it is that I do."

[nostalgebraist]

the scaling “inconsistency”: openAI’s new insight

I’ve now read the new OpenAI scaling laws paper.  Also, yesterday I attended a fun and informative lecture/discussion with one of the authors.

While the topic is on my mind, I should probably jot down some of my thoughts.

This post is mostly about what the new paper says about the “inconsistency” brought up in their previous paper.

The new paper has a new argument on this topic, which is intuitive and appealing, and suggests that the current scaling trend will indeed “switch over” soon to a new one where dataset size, not model size, is the active constraint on performance.  Most of this post is an attempt to explain and better understand this argument.

——

The new paper is mainly about extending the scaling laws from their earlier paper to new modalities.

In that paper, they found scaling laws for transformers trained autoregressively on text data.  The new paper finds the same patterns in the scaling behavior of transformers trained autoregressively on images, math problems, etc.

So the laws aren’t telling us something about the distribution of text data, but about something more fundamental.  That’s cool.

They also have a new, very intuitive hypothesis for what’s going on with the “scaling inconsistency” they described in the previous paper – the one I made a big deal about at the time.  So that’s the part I’m most excited to discuss.

I’m going to give a long explanation of it, way longer than the relevant part of their paper.  Some of this is original to me, all errors are mine, all the usual caveats.

——

1. L(​C) and L(D)

To recap: the “inconsistency” is between two scaling laws:

The law for the best you can do, given a fixed compute budget. This is L(​C), sometimes called L(C_min).  L is the loss (lower = better), C is your compute budget.

The law for the best you can do, given a fixed dataset size. This is L(D), where D is the number of examples (say, tokens) in the dataset.

Once you reach a certain level of compute, these two laws contradict each other.

I’ll take some time to unpack that here, as it’s not immediately obvious the two can even be compared to one another – one is a function of compute, the other of data.

2. C sets E, and E bounds D

Budget tradeoffs

Given a compute budget C, you can derive the optimal way to spend it on different things.  Roughly, you are trading off between two ways to spend compute:

Use C to buy “N”: Training a bigger model – “N” here is model size

Use C to buy “S”: Training for more steps “S” (gradient updates)

The relationship between S (steps) and D (dataset size) is a little subtle, for several reasons.

From step count to update count

For one thing, each single “step” is an update on the information from more than one data point.  Specifically, a step updates on “B” different points – B is the batch size.

So the total number of data points processed during training is B times S.  The papers sometimes call this quantity “E” (number of examples), so I’ll call it that too.

From update count to data count

Now, when you train an ML model, you usually update on each data point more than once.  Typically, you’ll do one pass over the full dataset (updating on each point as you go along), then you’ll go back and do a second full pass, and then a third, etc.  These passes are called “epochs.”

If you’re doing things this way, then for every point in the data, you get (number of epochs) updates out of it.  So

E = (number of epochs) * D.  

Some training routines don’t visit every point the exact same number of times – there’s nothing forcing you to do that.  Still, for any training procedure, we can look at the quantity E / D.

This would be the number of epochs, if you’re doing epochs.  For a generic training routine, you can can think of E / D as the “effective number of epochs”: the average number of times we visit each point, which may not be an integer.

Generally, E ≠ D, but we always have E≥D.  You can’t do fewer than one epoch; you can’t visit the average point less than once.

This is just a matter of definitions – it’s what “dataset size” means.  If you say you’re training on a million examples, but you only update on 100 individual examples, then you simply aren’t “training on a million examples.”

3. The inconsistency

L(D): information

OpenAI derives a scaling law called L(D).  This law is the best you could possibly do – even with arbitrarily large compute/models – if you are only allowed to train on D data points.

No matter how good your model is, there is only so much it can learn from a finite sample.  L(D) quantifies this intuitive fact (if the model is an autoregressive transformer).

L(​C): budgeting

OpenAI also derives another a scaling law called L(​C).  This is the best you can do with compute C, if you spend it optimally.

What does optimal spending look like?  Remember, you can spend a unit of compute on 

a bigger model (N), or 

training the same model for longer (S)

(Sidenote: you can also spend on bigger batches B.  But – to simplify a long, complicated story – it turns out that there are really just 2 independent knobs to tune among the 3 variables (B, N, S), and OpenAI frames the problem as tuning (N, S) with B already “factored out.”)

In the compute regime we are currently in, making the model bigger is way more effective than taking more steps.

This was one of the punchlines of the first of these two papers: the usual strategy, where you pick a model and then train it until it’s as good as it can get, is actually a suboptimal use of compute.  If you have enough compute to train the model for that long (“until convergence”), then you have enough compute to train a bigger model for fewer steps, and that is a better choice.

This is kind of counterintuitive!  It means that you should stop training your model before it stops getting better.  (“Early stopping” means training your model until it stops getting better, so this is sort of “extra-early stopping.”)  It’s not that those extra steps wouldn’t help – it’s that, if you are capable of doing them, then you are also capable of doing something else that is better.

Here’s something cool: in Appendix B.2 of the first paper, they actually quantify exactly how much performance you should sacrifice this way.  Turns out you should always stop at a test loss about 10% higher than what your model could asymptotically achieve.  (This will be relevant later, BTW.)

Anyway, OpenAI derives the optimal way to manage the tradeoff between N and S.  Using this optimal plan, you can derive L(​C) – the test loss you can achieve with compute C, if you allocate it optimally.

N goes up fast, S goes up slowly…

The optimal plan spends most incremental units of compute on bigger models (N).  It spends very little on more steps (S).

The amount it spends on batch size (B) is somewhere in between, but still small enough that the product E = B*S grows slowly.

But remember, we know a relationship between E and “D,” dataset size.  E can’t possibly be smaller than D.

So when your optimal plan chooses its B and its S, it has expressed an opinion about how big its training dataset is.

The dataset could be smaller than B*S, if we’re doing many (effective) epochs over it.  But it can’t be any bigger than B*S: you can’t do fewer than one epoch.

… and you claim to achieve the impossible

L(​C), the loss with optimally allocated C, goes down very quickly as C grows.  Meanwhile, the dataset you’re training with that compute stays almost the same size.

But there’s a minimum loss, L(D), you can possibly achieve with D data points.

The compute-optimal plan claims “by training on at most B*S data points, with model size N, I can achieve loss L(​C).”

The information bound says “if you train on at most B*S data points, your loss can’t get any lower than the function L(D), evaluated at D = B*S.”

Eventually, with enough compute, the L(​C) of the compute-optimal plan is lower than the L(D) of the dataset used by that same plan.

That is, even if the compute-optimal model is only training for a single epoch, it is claiming to extract more value that epoch than any model could ever achieve, given any number of epochs.

That’s the inconsistency.

4. The resolution

In the new paper, there’s an intuitive hypothesis for what’s going on here.  I don’t think it really needs the multimodal results to motivate it – it’s a hypothesis that could have been conceived earlier on, but just wasn’t.

Bigger models extract a resource faster

The idea is this.  As models get bigger, they get more update-efficient: each time they update on a data point, they get more out of it.  You have to train them for fewer (effective) epochs, all else being equal.

This fact drives the choice to scale up the model, rather than scaling up steps.  Scaling up the model makes your steps more valuable, so when you choose to scale the model rather than the steps, it’s almost like you’re getting more steps anyway.  (More “step-power,” or something.)

The resource is finite

Each data point has some information which a model can learn from it.  Finite models, trained for a finite amount of time, will miss out on some of this information.

You can think about the total extractable information in a data point by thinking about what an infinitely big model, trained forever, would eventually learn from that point.  It would extract all the information – which is more than a lesser model could extract, but still finite.  (A single data point doesn’t contain all the information in the universe.)

This is literally the definition of L(D): what an infinitely big model, trained forever, could learn from D separate data points.  L(D) quantifies the total extractable information of those points.

(More precisely, the total extractable information is the gap between L(D) and the loss achieved by a maximally ignorant model, or something like that.)

Converging in the very first step

As models get bigger, they extract more information per update.  That is, each time they see a data point, they extract a larger fraction of its total extractable information.

Eventually, your models are getting most of that information the very first time they see the data point.  The “most” in that sentence gets closer and closer to 100%, asymptotically.

How does this relate to optimal compute allocation?

The logic of the “optimal compute plan” is as follows:

Your model is an imperfect resource extractor: it only gets some of the resources locked up in a data point from the first update.  So you could extract more by running for more steps … 

…  but if you have the compute for that, you can also spend it by making your steps more efficient.  And, in the current compute regime, that’s the smarter choice.

It’s smarter by a specific, uniform proportion.  Remember, you should stop training when your loss is 10% higher than the converged loss of the same model.  If the converged loss is L, you should stop at 1.1*L.

Can you always do that?  If your model is efficient enough, you can’t!  As the first epoch gets closer to 100% efficient, the loss after the first epoch gets arbitrarily close to the converged loss.  Your loss goes under 1.1*L by the end of the first epoch.

At this point, the story justifying the L(​C) law breaks down.

The L(​C) law goes as fast is it does because upgrading the efficiency of your extractor is cheaper – in terms of compute spent per unit of resource extracted – than actually running the extractor.

This works as long as your extractor is inefficient.  But you can’t push efficiency above 100%.  Eventually, the only way to extract more is to actually run the damn thing.

Getting a bigger quarry

When you’re extracting a resource, there’s a difference between “improve the extractor” and “get a bigger quarry.”

If your quarry has 100 resource units in it, the strategy of “improving the extractor” can never get you more than 100 units.  It can get them to you faster, but if you want more than 100 units, you have to get a bigger quarry.

“N” sets the efficiency of the extractor.  “S” sets … well, it doesn’t exactly set the size of the quarry (that’s D).  There is an ambiguity in the S: it could mean running for more epochs on the same data, or it could mean getting more data.

But S does, at least, set an upper bound on the size of the quarry, D.  (Via D≤E and E = B*S, with B set optimally as always.)

With high enough compute (and thus model size), you’ve pushed the “extractor upgrades are cheap” lifehack as far as it can go.  With this efficient extractor, taking S steps (thus making E = B*S updates) sucks up most of the information theoretically extractable from E individual data points.

The learning curve L(E) of your model, as it makes its first pass over the dataset, starts to merge with L(D), the theoretical optimum achievable with that same dataset.  You trace out L(D) as you train, and the relevant constraint on your performance is the maximum data size D you can obtain and train on.

Where we are now

In the compute regime that spans GPT-2 and the smaller variants of GPT-3, extraction is far less than maximally efficient.  The L(​C) strategy applies, and the smart move is to spend compute mostly on model size.  So you make GPT-2, and then GPT-3.

Once we get to the full GPT-3, though, the extractor is efficient enough that the justification for L(​C) has broken down, and the learning curve L(E) over the first epoch looks like L(D).

Here is that as a picture, from the new paper:

The yellowest, lowest learning curve is the full GPT-3.  (The biggest GPT-2 is one of the green-ish lines.)  The black line is L(D), maximally efficient extraction.

You can see the whole story in this picture.  If you’re in one of the smaller-model learning curves, running for more steps on more data will get you nowhere near to the total extractable info in that data.  It’s a better use of your compute to move downwards, toward the learning curve of a bigger model.  That’s the L(​C) story.

If the L(​C) story went on forever, the curves would get steeper and steeper.  Somewhere a little beyond GPT-3, they would be steeper than L(D).  They would cross L(D), and we’d be learning more than L(D) says is theoretically present in the data.

According to the story above, that won’t happen.  We’ll just converge ever closer to L(D).  To push loss further downward, we need more data.

Implications

Since people are talking about bitter lessons a lot these days, I should make the following explicit: none of this means “the scaling hypothesis is false,” or anything like that.

It just suggests the relevant variable to scale with compute will switch: we’ll spent less of our marginal compute on bigger models, and more of it on bigger data.

That said, if the above is true (which it may not be), it does suggest that scaling transformers on text alone will not continue productively much past GPT-3.

The GPT-3 paper says its choices were guided by the “grow N, not S” heuristic behind the L(​C) curve:

Based on the analysis in Scaling Laws For Neural Language Models [KMH+20] we train much larger models on many fewer tokens than is typical.

(“KMH+20″ is the first of the two scaling papers discussed here.)  Even following this heuristic, they still picked a huge dataset, by human standards for text datasets.

In the above terms, their “E” was 300 billion tokens and their “D” was ~238 tokens, since they updated multiple times on some tokens (cf. Table 2.2 in the GPT-3 paper).  The whole of Common Crawl is 410 billion tokens, and Common Crawl might as well be “all the text in the universe” from the vantage point of you and me.

So, there’s room to scale D up somewhat further than they did with GPT-3, but not many orders of magnitude more.  To me, this suggests that an intuitively “smarter” GPT-4 would need to get its smartness from being multimodal, as we really can’t go much further with just text.

[an-apocalypse-of-magpies]

On the world of Mortal Engines, class, and the metaphors of consumption

This is less an essay and more a collection of thoughts. Basically I just saw a video on the Mortal Engines film and its being a civilisation too stupid to exist. I got fed up, mainly because so many of the criticisms amounted to ‘the book did it better’ with little elaboration but also the arrogantly grating voice of the presenter got on my nerves, but I cannot deny the points made and in fact wanted to elaborate further on the worldbuilding of this series and, while unrealistic, look at why the books were so engaging.

Some background to start off - Mortal Engines is a four-book series (and three-book prequel sub-series) written by English author Phillip Reeve, and depicts a bleak post-apocalyptic world. North America is uninhabitable and lost to the sands of time, irradiated, poisoned, and flattened by war. Eurasia is mostly barren plains. And, of course, the central premise - towns and cities have raised themselves onto mobile platforms and trundle about. Well, mostly. A major antagonist to this system is the Anti-Traction League, a collective of nations hiding out in old east China, the Indian subcontinent, southeast Asia and some of Africa. They are seen as barbarians and heathens by much of the world for refusing to mobilise, instead hiding in stationary citadels behind their mountains. The Traction Cities near-universally engage in a philosophy of Municipal Darwinism, a savage system of bastardised pseudo-biology where cities literally predate each other and ‘consume’ each other for resources. Cities eat towns, towns eat smaller towns. Some towns and cities deliberately adapt to cheat the system and make themselves a less appetising target, or for that matter a more aggressive and efficient hunter.

THE TRACTION CITIES

The first three books tend to focus their action on one or two cities, whereas the last is a bit more of a road trip. The other consistent thread is multiple characters’ stories running concurrently, usually reconnecting near the end. This allows the books maintain an open, almost global scale - you’ll nearly never not be moving, even sitting still on a city, which reinforces the theme of unnatural life. The first book focuses on London, which has been sulking in what was once Britain (by sheer happenstance on their part and pure irony on ours), and is suddenly running at full pelt back into Europe and eastward as fast as her engines can carry her. Why? London’s not the biggest city around, and the vast expanse of Eurasia is now the Great Hunting Ground - it’s where the big boys play, and by play I mean ‘savagely predate each other’. It’s dangerous territory for a little city. But over the first book, it becomes increasingly apparent that Traction Cities are increasingly non-viable option for existence. Fuel is scarce, prey moreso, and what morsels London can confidently snap up will not sustain it for long. There is an ecosystem at play here - static settlements can farm resources, but are universally seen as food, either by small bandit settlements to raid for supplies or for larger towns to just straight-up eat. Small towns too small to hunt tend to be miners or gatherers, either mining minerals to use or trade, or gathering resources like wood from natural deposits or sifting through the waste heaps left by bigger cities. Most cities bigger than that are ‘urbivores’, or hunter towns, that hunt and eat smaller prey or opportunistically scavenge the ‘carcasses’ of dead cities. I mentioned specialisation earlier, and like in nature, species and cities can occupy a niche that gives them an advantage and thus increased chance at survival. Airhaven, for example, is a politically-neutral city in the air that floats around Eurasia seasonally and serves as a rest stop, fuelling station and trading exchange for airship pilots the world over, Tractionist or no. Tunbridge Wheels is a pirate-run town that has a lightweight wooden chassis and flotation devices to hunt amphibiously in a world where many small towns escape threat by setting up on islands.  Panzerstadt-Bayreuth is a conurbation of four massive cities, too big to survive long without prey, they banded together to take down the biggest of prey (it’s unclear whether they achieve this through sheer size or whether they decouple and become a pack hunter). Anchorage, the last American city, neutered its own jaws to increase mobility, skating around the frozen north too fast for threats to catch up with, and survives on trade. Brighton is a pleasure city that paddles around the warm Mediterranean, technically still a predator but with no real agenda and about the only city left that can be called a tourist city (it’s run on the back of brutal slave labour). And these are just the major ones. Throughout the books, cities are treated like living things ... like mortal engines.

And like living things, they need resources to survive.

A DYING WAY OF LIFE

The books are inconsistent on the origins of Traction Cities, as it turns out deliberately - history is written by the winners, after all. But it’s all closely tied to the ‘apocalypse’ part of the post-apocalytic I mentioned earlier. Long ago in-universe, long into our future, was a terrible event known as the Sixty Minute War. This war tore the world asunder with nuclear and quantum energy weaponry. America, the epicentre, is simply no more (it turns out there are some fertile areas in Nova Scotia, but for the most part America is dead). Entire new mountain ranges were born, notably the Tannhäusers in East Asia that shield the heartland of the Anti-Traction League. There was a long period of geological and tectonic instability. According to legend, Traction Cities arose to escape these instabilities. In other words, like animals will flee a volcanic eruption, cities first became mobile to escape and survive. Trade was likely facilitated by towns literally being able to park next to each other. Ironically, London was also where everything changed. After Nikola Quercus conquered (static) London with his mobile fortresses, he decided to upgrade and raise London onto wheels to become the first fully-mobile city. And he did it for war. After all, there’s no better comeback to ‘you and what army’ then literally rolling up with your entire city. By the series present, the idea had caught on and grown into the ideology described above. But herein lies the problem. Early Traction London was a tiny little thing. Now it’s not even the biggest fish in the pond, but it’s still HUGE. And, as we all know, big things need lots of energy to go. London is described as having a top speed of about sixty miles per hour at the height of a hunt. So, you need fuel. There is still oil in this world, mainly because they now have no qualms about mining Antarctica, but if you think there’s nearly enough crude oil to run a world full of cities like London you are sorely mistaken. Wood’s not much better off. And, of course, Traction Cities tend to run on some form of internal combustion engine - it’s only at the very end of the traction era that science has advanced enough for a town to experiment with magnetic levitation. So what do they burn? Well, bits of other prey towns. Do you see the problem? Use fuel to hunt towns, burn those towns for fuel. What next? And it’s not just fuel. London captures a little salt-mining town called Salthook at the beginning of the first book to introduce us to the concepts at play, and we see what goes on in the Dismantling Yards - part of a system literally called the Gut, in case the metaphor wasn’t clear yet. Everything is recycled. Bricks, mortar, steel, wood, everything. Because the state of technology is so weird in this world, Old-Tech (technology from before the SMW) can be incredibly valuable to history and/or science, and London is keen to snaffle that up too. The people are interred into refugee camps, though if you know anything about how real-life Britain treats refugees you can probably see where that is going. And it’s not enough. It’s never enough. Food is an even more pressing concern. Unless you’re very rich (more on that in a mo), food is mostly algae-based, then hardy vegetables that grow quickly like cabbage. And it’s running out fast. And London’s a big city with a lot of resources at its disposal. Most cities don’t even have that. A lot of cities are starving on the wheels, city and populace alike. A lot of cities run on slave labour, and feed those slaves as little as they can get away with. Shan Guo, home of the Anti-Traction League, is a green and vibrant land only because it doesn’t have cities running over or eating its farmlands every other day (and, again, city folk generally don’t know this - they’re given endless propaganda that Anti-Tractionists are barbarian warbands a la Mad Max). A lot of the A story is told from the point of view of Tom Natsworthy, who until the events of the book had never left London. He’s never seen bare earth or walked on mud before. He’s never seen a horse. The idea that you can survive, much less thrive, outside of a Traction City is alien to him. But on the city he came from, everything is rapidly running out, and some cities are turning to desperate measures to survive, including Arkangel openly bribing pilots to sell out the locations and courses of nearby cities. A chilling scene in the first book even has Tom see, from the safety of the air, the corpse of Motoropolis, a city not unlike London that literally just starved to death, running out of fuel and helpless as the scavengers closed in. It’s been weeks since the city stopped, and the narrative description evokes the grotesqueness and sadness of a whale carcass. Sheer Jingoism is about the only thing keeping Municipal Darwinism alive - Traction good, stationary bad.

CLASS, CLASSISM, AND OTHER SOCIAL OPPRESSIONS

In a world so starved as this, compassion is hard to come by. Cities still exist mainly by virtue of rigid social stratification, and often that stratification is literal - most medium-to-large cities have tiers, and will generally arrange those tiers based on social class. London, for example, has seven tiers. The bottom two tiers are dominated by the Gut, the engines, and homes and communities of the workers who keep them running. Tiers 4 and 3 are miscellaneous proles of increasing social standing. Tier 2 is mostly what I’d call ‘tourist London’ - lots of the nice bits and the establishments that London likes to be proud of. Because of his work at the London Museum, this is the quality of life Tom Natsworthy was most used to. Tier 1 is High London, where all the rich live and have their amenities and nice parks (and even that doesn’t last - London’s food shortage means even the High London parks are eventually, begrudgingly, turned over for food production). Katherine Valentine, the hero of the first book’s B plot, lives here. Finally there’s Top Tier, which is purely administrative. The only buildings are the Guildhall (the seat of government), St Paul’s Cathedral (which the Engineers’ Guild have secretly been installing a deadly superweapon in under the guise of ‘restoration’ work) and the headquarters of the Guild of Engineers, the most powerful of London’s Guilds. Social stratification is nearly non-existant, and people are shown to get very uncomfortable when out of ‘their space’. Tom is sent to work in the Gut during the capture of Salthook as a punishment before the plot ejects him from London, and he notes being actively intimidated by the claustrophobia, the dirt, the rough and burly labourers, and the noise. But despite Tom’s relatively privileged life - he lives near High London, above the heat and noise and smoke of the engines, in the care of one of the top four Guilds of London - he is of very low social status. Tom Natsworthy is an orphan; his parents were Historians, but were killed when an accident occurred and part of Tier 3 collapsed, crushing anything on Tier 4 beneath. Even before that, the Natsworthys were middle class at best, but being orphaned meant being left to the care of an orphanage run by the Guild of his parents, the Historians. The Historians were Tom’s only source of education, and eventually they would employ him, but with no parents or money, Tom can only afford a Third-Class apprenticeship. He has no upwards mobility within the Guild, and with no money he can’t leave and train with another. His dream of being a pilot trader, or better yet adventurer, will never come true under normal circumstances. The rich live in a completely different world yet. Katherine Valentine, daughter of the Head Historian and the Lord Mayor’s ‘right-hand man’ Thaddeus Valentine, has a positively bougie lifestyle with not a care in the world. Ironically, though, it is through Katherine’s eyes that the horrors of London’s class system are revealed. Trying to find information about her father’s would-be killer, Katherine finds herself regularly travelling to the Gut, eventually befriending an apprentice Engineer who witnessed the attack. But in the Gut, life is very different. It’s not just a life of hard labour and smoke - petty criminals and the aforementioned ‘refugees’ are tasked with working dangerous and sickening jobs like managing the city’s sewage. And by that, I mean ‘harvesting literal faeces to be converted into food and fuel’. The foreman overseeing their work admits they feed such criminals nothing else. And he has the gall to be annoyed that they keep dying of diseases like cholera and typhoid! These people are denied medical care, denied treatment, denied even basic food other than being told to literally eat sh*t. And when they inevitably die? They get sent to the Engineerium to be turned into robotic zombies that can never get sick, tired or unhappy. And, eventually, they’ll be put right back to work. The crimes these criminals did to deserve this, remember, include petty theft, criticising the Lord Mayor, and living aboard a town that got eaten. The foreman literally cannot fathom why Katherine would care about these people’s wellbeing - after all, they’re just criminals. The Engineerium’s end goal in all this is, again, to staff the entire lower tiers with robot zombie workers who will never grow tired, get sick, complain or protest their lot in life, and will never disobey orders, and just enough human overseers to keep things running smoothly ... because that’s what these people are worth to London, cheap, unending labour. Katherine can’t even bring herself to tell her high-class peers about what she learned down there, because it’s such a different world that they would never empathise, much less care. Again, slave labour is common in this world, especially child slavery - Brighton runs on it to maintain its image as a floating Caligula’s Palace, and in Arkangel slavery is so normal that we watch a rich man beat a slave nearly to death for the crime of bumping into him. In the second book, we see the logical end-point of this. Anchorage’s social structure has completely fallen apart due to a plague in recent years that turned to once-proud ice city into a ghost town manned only by a skeleton crew. The margravine, Freya, is only 14, but with her parents dead, she finds herself in charge of the whole city. She has no household staff, apart from Smew, who finds himself constantly juggling outfits to adopts the roles of steward, chamberlain and so on. His official role before the plague was ... erm ... the Dwarf. He was there in a manner similar to a court jester, for the amusement of the margrave due to being a little person. But the head navigator is just ... the woman who kept the maps. The head engineer is going half-mad, seeing his dead son staring at him from the shadows, and the only reason the town’s still going is because his systems are the best on the ice and can mostly run on automatic. They have no doctor. The only other people of consequence in Anchorage are the Aakiuqs, the Inuit couple who run the air-harbour. The common workers of Anchorage number in the mere dozens. And yet, because they’re so fixated on their traditions, nobody will drop the formalities and just admits that they’re trying to uphold a class system that doesn’t work anymore. No, that’s not quite right - everybody realises it’s pointless to maintain the artifice of Anchorage’s social heirarchy, but nobody wants to be the first one to say it out loud. Much like Municipal Darwinism, nobody want to address the elephant in the room, that the system is broken and that people hold onto it because it’s comfortable in the face of uncertainty. Only in Anchorage’s darkest hour, when everything has been turned upside down and the conquerors are on their doorsteps, do the agree to drop the formalities, drop the artifice of class, and address each other as people, say what they think, and work to save what they have left. And of course, there’s the racism in the world. Life on mobile cities has made cultures smaller and more insular, considering we mainly see this series from the point of view of culturally-English towns. Throughout the first book there is a clear west vs east divide - the Traction Cities are generally English-speaking or multicultural enough that English will get you by. The Anti-Tractionist League, meanwhile, are south or east Asian, or else African, and are commonly understood to be ‘those brown people’. The only ethnically white Anti-Tractionists are from ‘Spitzbergen’ (likely Scandinavia/Finland and northwest Russia) and Hester Shaw’s family, and the latter lived on a town that floated out to an island and gave up running from predators forever. The way Tom reacts to this attitude calls to mind the way racists might refer to ‘race traitors’. There’s even an in-universe slur for people who live in static settlements; ‘Mossies’, because ‘a rolling town gathers no moss’. However, when Tom is taken to Shan Guo itself, he realises that all the propaganda he’d been fed his whole like is exactly that - propaganda. Shan Guo is described as beautiful - an endless patchwork of rolling fields and farms, colourful, bright, vibrant, heaving with life and energy. The Anti-Tractionists aren’t vicious savages, they’re just ... people. Tom can’t understand it at first. He wonders how people can live without the hum of engines or the vibrations of deckplates - he subconsciously equates city life with, well, life, and the absence of that makes him uneasy. But he can also see this culture before him, thousands of years old, outlasting even the end of the world, and he realises there is another way. The next time he sees London, he sees it from outside, from the side of the hunted, and he realises it’s not beautiful or efficient, just dirty, and huge, wrapped in its own waste smoke and driven only by destruction. For the rest of the series, even with the rise of the radicalised Green Storm (Anti-Tractionists Lv2), large Traction Cities are consistently the enemy. Tractionism as a culture is understood to only represent imperialism, destruction, and consumption, literally and figuratively.

SCIENCES SANS FRONTIERES

It should be noted that science and technology are not universally reviled by the series. As a dieselpunk series, a certain degree of technology is fundamental to the series existence. But this is a very different world than the one we know. On the one hand, engines exist that can drive entire cities. On the other, computers basically do not exist. The rare few that still exist are not in working condition, and nobody knows how to restore them. Heavier-than-aircraft don’t really exist - the third book introduces some, but they’re small, experimental ... barely more than short-range toys designed for flashy air shows but not real travel. The main form of personal locomotion in this world is by airship, and this world’s airships are far beyond anything we’ve made in our time. But lost technologies are heavily associated with the hubris and destructiveness of the Ancients. Until now. Like I said, the most powerful Guild in London is the Engineers’ Guild. And they got that way under the leadership of now-Lord Mayor Magnus Crome. It should be noted that Crome genuinely loves his city and wants it to survive no matter the cost. But under Crome, the Engineers began to dabble in sciences considered unethical to downright taboo. Most notable is the MEDUSA Project. Through Thaddeus Valentine, London came into possession of an energy weapon from the SMW ... and, more importantly, the working computer that runs the thing. In terms of Darwinist Evolution, this is like giving a monkey a gun and teaching it how to use it. MEDUSA exhibits a level of power no other force on Earth can match, and London is forced to deploy it early in a crisis. Originally, the plan was to march up to Batmunkh Gompa, the Shield-Wall that represents the only break in the mountains around Shan Guo big enough to permit a city, and blast it to cinders. Unfortunately, London attracts the attention of a bigger, hungrier city about halfway there, and is forced to fire MEDUSA at it to save its own skin. The sheer terror of what that weapon represents is revealed then. Panzerstadt-Bayreuth was the fusion of four massive cities, each one bigger and more powerful than London. MEDUSA killed it dead in one stroke - the energy beam set the entire city ablaze and ignited its fuel stores. Her engines nearly immediately exploded. When the fires go down enough for an Engineer scout ship to investigate, the people had been almost flashed into glass. The flash of light from the attack is so bright that, hundreds of miles to the south, Tom and Hester see the sky light up like a new dawn. The people of London are relieved, of course, that they didn’t all die that night, but more than that the entire city become suffused with the excitement of just how easy it would be to kill ... well, anyone they like, really. London doesn’t even stop to devour Panzerstadt-Bayreuth, as the Engineers can’t afford for the Shield-Wall to prepare for their arrival. Appropriately, and karmically, the finale has an accident lock down the computer lock down, with MEDUSA unable to fire but unable to stop gathering energy, and London melts under the heat of MEDUSA’s glare. But that wasn’t the only scientific sin committed by London’s engineers. I’ve already mentioned London trying to repurpose faeces as food, but we need to talk more about the Stalkers. Stalkers are kinda like discount Cybermen from Doctor Who - dead bodies, threaded with weird old machines and coated in armour, their brains hooked up to simple computers. Originally conceived as soldiers, they were believed long dead. However, one survived to the modern by sheer survivor instinct - Shrike. Through negotiations that are not the purview of this essay, he allowed the Engineers of London to take him apart and figure out how he worked, and hoo boy they did. The Engineers figured out how to manufacture their own Stalkers. The first batch are used as law enforcement like the Worst Robocops, but, again, the plan was to have Stalker workers all over Low London. Katherine, learning this, likens it to London ‘being a city of the dead’ (Apprentice Engineer Pod, to whom she is talking, grimly notes that the Deep Gut Prison is so awful, so callous with human life, that it already feels like that). Logically, the end-point of this idea is to have all workers in London be the resurrected dead, with just enough living to keep things in order ... oh, and they’d all be loyal to the Engineers, because remember, no Freedom of Speech here, and you can be sent to do the worst form of prison labour for dissenting against the Lord Mayor. With Crome being both Lord Mayor and Head Engineer at once, the Engineers’ creed is as good as law - traditionally, London Lord Mayors forsook their former Guild allegiances to show their representation of all of London, and Crome’s refusal to do that caused a bit of a stir. The Engineers are also keen to arm their security teams with some form of energy pistols, despite guns being outlawed in London and the police are only allowed crossbows. Crome’s rationale is the same as every two-bit mad scientist villain, of course - that science should not be held back by moral restrictions, and that progress for progress’ sake is essential for London’s survival. Really, it’s the Engineer’s survival, as they’re rather loathe to share these advancements except to exert power on those around. London isn’t the only example of technology being used to leverage control and benefit the ruling classes. Grimsby is a sunken wreck of a city somewhere in the north Atlantic, yet due to a complex series of airlocks the interior of the city is a secret hideaway of the Lost Boys, a society of children stolen from aquatic towns and trained to be thieves under the watchful eye of the mysterious Uncle. They will then take submarine walkers, attach to passing towns, steal whatever tools, fuel, food and riches they can carry, and vanish back into the depths. Uncle, naturally, takes the lion’s share of the haul. But Uncle maintains his power by careful access to technology, only letting the Boys have what they need and juggling the power structure by choosing team leaders, and punishing insubordination harshly and publicly. Uncle sees and hears everything in Grimsby with his surveillance network, and can address any give Boy in a heartbeat, training the Boys to never expect privacy from him, so that when he demands a progress update from a mission, they never question him. He rewards Boys who do well on burglaries, but more importantly than that, he chooses team leaders according to apparently inscrutable whims. The Boys believe it’s a mark of favour from Uncle, and thus social status, to be trusted with the limpet command and all the tech that comes with. Really, Uncle carefully give command to people he can trust to remain loyal to him, even if that means passing over a more talented Boy who might get a bit uppity. Even in a more mundane way, higher status in the Lost Boys means you can move closer to the heart of Grimsby, where you’re less likely to wake up and find your bedroom wasn’t as watertight as you thought and flooded in the night. Uncle, naturally, doesn’t care if a few Boys drown, so long as he doesn’t lose anything useful. Technology, and in particular access to unusual technology, is the dimension on which power is really decided.

THE END OF AN ERA

We’ve already established that this world is not a sustainable one. There are only so many cities. The inherent entropy of Municipal Darwinism is really showing. Once upon a time, big cities could ‘reproduce’, creating little satellite towns that could grow and become independent - even London had some - but those are no more. In a greedy desperation to keep moving, the predators are not reproducing, and static settlements can’t spread and grow fast enough to count there. The attack of London, and MEDUSA, turned staunch opposition into outright war, with the Green Storm being willing to doublethink their way into using the weapons of the Traction Cities in their fight to stop the Traction Cities, even recruiting ex-London Engineers to make weapons and stalkers for them, and eventually even seeking out another ancient superweapon - an orbital laser called ODIN - without a hint of irony. The Green Storm eventually face internal resistance, from Anti-Tractionists who disagree with the outright terrorism angle, and eventually crumbles. The last great Traction Cities stop. The last mobile city is New London, no longer a hunter but a trade platform, and even that probably stopped hovering about at some point. The ending is told by the great survivor, Shrike, who has cheated Death again and again, who outlived Tom Natsworthy and Hester Shaw, Valentine, Magnus Crome, and a thousand other heroes and villains. When he awakes, long in the future, Traction Cities are not even ancient history. They’re a dream, a fantasy, too incredible to be true. But Shrike remembers, and he teaches people the story of London and Anchorage, Arkangel and Airhaven, Brighton and Harrowbarrow. Did they learn the right message from Shrike’s story? Did they learn that ruthless imperialism is like hunting faster than the food can come back, and that you will starve before you have everything you ever wanted? Did they learn that hoarding resources, gatekeeping knowledge, will lead to ruin? Did they learn, or will the repeat the same mistakes of the greed and gluttony of the Traction Era? Well, who knows.