[Read in full on NHAM]

Misaligned Aardvarks

Serialising the album Pretentious Incongruity by Key 13 (@keefmarshall)

Pretentious Incongruity: 2020-2024 is a retrospective compilation looking back at some key milestones in the first five years of Key 13 as a recording identity.

Track 11

Where to start with this one.. Firstly, I’m really not a fan, especially of the arrangement and mix, which I just can’t get right. I’m not sure this new mix is any better than the original – it might be worse. But it has to be here, it’s the one everyone knows, because somehow, against all odds, it won the 2024 Fedivision Song Contest!! How on earth did that happen. I still feel they gave me someone else’s votes.

It all came from a throwaway conversation between two contacts on Mastodon (inpc and elsemusic), who have probably been mystified about the whole thing ever since.

I had the opening section from around December of the previous year, up to when the guitar comes in – this came out of playing around with the presets in one of Cherry Audio’s Moog emulations (Memorymode, I think) – but had got stuck there for months. I had the main vocal line in my head but didn’t have any of the words, and didn’t know where the rest of the track was going – nor how the aardvarks got out of their conundrum! I did know, from the start, that it was going to be my Fedivision entry for the year – although I thought it would probably come in as an epic long track, and would need to be edited down to meet the 4:00 max length for the competition.

In the spring of 2024 I got fixated on getting out an EP called Sea · Sky · Clouds, remastering some earlier tracks of mine – I don’t know why (see: Monotropism, also referenced later on!) but I just couldn’t focus on anything else musical until I got this released. Finally on April 25th, I got it out. However, there was also the little matter of the Not Bonk What I Call Wave: Remixes album release which needed to go out on May 10th, and although others were doing most of the work, I still had to prepare the audio files and do the Faircamp release.

When the Fedivision call for entries came around, the submission period ran from 5th to the 12th May, but I had the above Bonk Wave release planned and was about to go on holiday, so I was cutting it a bit fine. I was also on a tough project at work which was sapping my energy – I was generally struggling a bit. In the end, the whole thing was completed in just one session, during the only spare time I had – a few hours, late into the evening, on the 6th May. I knew the mix was really bad, pretty much unlistenable, so somehow managed to find more time one evening midweek (getting myself into trouble as I should have been doing something else – e.g. packing for holiday!) to spend another hour or two fixing it up a little before finally getting it submitted on the 8th. Luckily (and hugely coincidentally!) it had come in at exactly the same length as my previous entry, well under 4 minutes, so no editing down required.

Given how rushed the whole process had been, I wasn’t holding out much hope. When the Fedivision entry list came out, not only were there 71 other entries (!) which was astonishing, but mine was literally number 72, last on the playlist. The chances of anyone even getting to it through 4 hours of music were pretty slim.

As the results were announced, from position 10 downwards, I prepared for the worst. When we got to number 2 and I still hadn’t been mentioned, I thought that was it. But somehow, miraculously, there it was.. number 1! I still have no idea how, or why – there were several objectively-better tracks there. So sorry to all the other entries, you have my deepest apologies.

Influences: There’s a lot going on – I was trying to capture the sound of some early Italian 70s progressive rock bands – especially Premiata Forneria Marconi, but the vocals are clearly styled after Yes. It’s been pointed out there’s some Marillion flavours as well, which is fair enough.

– From Key13’s Pretentious Incongruity: Track Notes

Listen

mirlo.space/widget/track/10886

Support Key 13

• Buy the album
• Follow Key 13

Lyrics

Welcome to the mesmerising tapestry of the African savannah.
The golden hues of the sunset cast a warm glow upon the vast landscape.
Zebras graze peacefully,
and giraffes’ graceful silhouettes traverse the horizon.
But something here is not quite right.
As you can see, the aardvarks are… misaligned.

Recklessly joyful oblivious elephants
Stampeded without a care
All meditating in deep contemplation
The aardvarks were caught unaware
And that is how they lost their way.

These sad creatures stare listlessly into the distance.
It’s hard to watch.
They stand confused and vacant:
Is there any hope left?

[Orycteropus afer]

Since the encounter with perilous pachyderms
Things took a turn for the worse.
Misaligned aardvarks are no laughing matter
It shattered their whole universe.
Now they must try to re-align.

[Orycteropus afer]

But wait! What’s this?
Is that… a shaman?!
Extraordinary!
He’s collecting them together.
They seem to be starting to chant!

Orycteropus
Orycteropus
Orycteropus
Orycteropus afer!

It’s working!

… and now the aardvarks are aligned.

#PretentiousIncongruity #serialisation

Okay, Back of the napkin math:
- There are probably 100 million sites and 1.5 billion pages worth indexing in a #search engine
- It takes about 1TB to #index 30 million pages.
- We only care about text on a page.

I define a page as worth indexing if:
- It is not a FAANG site
- It has at least one referrer (no DD Web)
- It's active

So, this means we need 40TB of fast data to make a good index for the internet. That's not "runs locally" sized, but it is nonprofit sized.

My size assumptions are basically as follows:
- #URL
- #TFIDF information
- Text #Embeddings
- Snippet

We can store an index for 30kb. So, for 40TB we can store an full internet index. That's about $500 in storage.

Access time becomes a problem. TFIDF for the whole internet can easily fit in ram. Even with #quantized embeddings, you can only fit 2 million per GB in ram.

Assuming you had enough RAM it could be fast: TF-IDF to get 100 million candidated, #FAISS to sort those, load snippets dynamically, potentially modify rank by referers etc.

6 128 MG #Framework #desktops each with 5tb HDs (plus one raspberry pi to sort the final condidates from the six machines) is enough to replace #Google. That's about $15k.

In two to three years this will be doable on a single machine for around $3k.

By the end of the decade it should be able to be run as an app on a powerful desktop

Three years after that it can run on a #laptop.

Three years after that it can run on a #cellphone.

By #2040 it's a background process on your cellphone.