AIs Named by AIs | Hacker News Reader

AIs Named by AIs

aiweirdness.com

133 pointsbryanrasmussen22 comments

Open articleView on HN

My favorite Culture Ship name to date remains the "Frank Exchange Of Views".

I've had a good time this morning (re-)reading https://en.wikipedia.org/wiki/List_of_spacecraft_in_the_Cult....

You might enjoy @cultureshipnames on Twitter.

https://twitter.com/cultureshipname

A name only made better by the ship class, "Psychopath"!

Talking of which. I really love how the class names of the various warships are, lets say, less than subtle. Something of a self referential reminder that in war, one must accept that you're always the villain of someone's story, and war is never pretty.

Yes. War is obscene to the majority of the Culture, who are (somehow) a hyper-left, atheist civilization of organics and superAIs. They tag Elua hard.

They struggle with the lofty, unyieldy overhead of that ethos, but they seem to get by with insanely advanced technology and a sense of humor.

https://en.wikipedia.org/wiki/The_Culture

How much of this is the AI doing work and how much of it is the author cherrypicking 20 names out of 1000? I still think it's remarkably impressive that an AI can do this, but it's good to put it into perspective.

GPT-2 generated text in general has much higher signal-to-noise ratio on output. (about 10% are good, which is much, much better than the 1% with RNN approaches)

Holy shit those are close to culture names. Weeeeird. I wonder how Banks thought up the names for his ships? He did a good enough job of it that the Halo writers were inspired by it and followed in the same vein. It's a departure - before Banks, sci fi authors (think star trek for example) just followed navy naming conventions.

I really would like to learn enough to play with these tools as the author as, sounds great fun.

The Colab notebook linked in the article (which I made) is designed to make it as user-friendly as possible to finetune GPT-2 and generate text from it. Unfortunately "as possible" is load-bearing in this case, as GPT-2 is slightly more difficult to work with.

In the case of short-form content like this, using a single-column CSV with each document as a row will automatically prepend/append the texts with appropriate start/stop tokens, which you can then force the model to generate text within those bounds and regex out the text you want. (this is what gpt-2-simple does with the prefix, truncate, and include_prefix parameters to generate())

Ah I loved the ship names of Halo lore.

Truth And Reconciliation, Undiminished Entelechy, Pillar Of Autumn, Forward Unto Dawn, Say My Name, In Amber Clad, Ready Or Not, Glasgow Kiss, Do You Feel Lucky, Tripping Light...

https://www.halopedia.org/UNSC_Navy#Vessels

Every time I stumble upon AI generated texts it makes me feel weirdly scared. It's like reading texts written by someone having a stroke.

Also, Skynet confirmed.

> Hand Me The Gun And Ask Me Again

> Protip: Don’t Ask

Hand Me The Gun And Ask Me Again (and the whole first list on the page) are examples of names from The Culture, not actually AI generated. Protip: Don't Ask fits it nicely though!

https://en.wikipedia.org/wiki/List_of_spacecraft_in_the_Cult...

Semi-related fun fact - SpaceX's landing barge names are taken from these: Just Read the Instructions, Of Course I Still Love You, and (under construction) A Shortfall of Gravitas.

Neural nets have also been fed training sets of climbing route names, from various areas (Boulder, Joshua Tree, and UK).

UKC article here: https://www.ukclimbing.com/articles/features/neural_network_...

I'm not sure that I'd want to find myself on "Ramp of Lies" !

Just FYI, if you follow the links, it tracks back to the same blog. As do many/most internet posts about "weird AI names for things", at least in my experience.

https://aiweirdness.com/post/174885047997/rock-climbing-rout...

I imagine the Ramp of Lies is so named because nearly every hold is deceptive. Many seemingly decent crimpers are just tricks of light and shadow, the jugs are all remarkably less positive than they look, and even the problem itself seems slabby but is actually on a gravity hill.

Love the connection made by the author to the Culture novels - I’m reading one right now and they are great fun

“Dog food made of maggots (Malgobias & Mealybugs).

Cucumber (Spinosadaceae & Chlorophyllaceae).

Rabbit meat made from rabbits' excrement.

Rabbit eggs eaten raw in Japanese restaurants.

Mangrove shellfish made from the carcasses of mountain sheep, or the stumps of trees.

Hivebeast (Lamiasis & Mealybugs), or live fish eaten raw in Japanese restaurants.

Buckwheat bread in Japanese restaurants made entirely of boiled milk.

Rabbit fat served as a dessert by Koreans and many Japanese visitors.

Meat from rabbits eaten in the wild. These include the skins of rabbits, the intestines, and the ears.

Pork meat eaten raw in Japanese cafes.

Honey (Acacia chrysocystis).

Dried mushrooms (Dioecious mushrooms).

Dried mushrooms grown on hot rocks (Tepidolites), or ground on a pile of sand on a hot day.

Fresh meat from wild rabbits.

Ditch dumplings made of beef (Chihuahua, Mexican, Japanese).

Jinbu cake made with fresh vegetables.

Gourd cakes made with meat served as a dessert.”

This blog is tons of fun.

Hand me the gun and then ask me again is a great name for an album

"Computer, what should I name my cat?"

"Kill All Humans".

Okie dokie.

Is there any way to do something like this without any programming? Giving an AI a list of input and generating names?

These remind me of card names for a trading card game.

"Hacker News is dedicated to uncovering the inner workings of The Dark Net, a.k.a. the place where your kids (or anyone else) can find something they don't want. Join us for stories on the Dark Net, where it all started, and everything in between."

Linked from article: https://talktotransformer.com/

Reply on news.ycombinator.com