Man I feel this, particularly the sudden shutting down of data access because all the platforms want OpenAI money. I spent three years building a tool that pulled follower relation data from Twitter and exponentially crawled it’s way outwards from a few seed accounts to millions of users. Using that data it was able to make a compressed summary network, identify community structures, give names to the communities based on words in user profiles, and then use sampled tweet data to tell us the extent to which different communities interacted.

I spent 8 months in ethics committees to get approval to do it, I got a prototype working, but rather than just publish I wanted to make it accessible to the academic community so I spent even more time building an interface, making it user friendly, improving performance, making it more stable etc.

I wanted to ensure that when we published our results I could also say “here is this method we’ve developed, and here you can test it and use it too for free, even if you don’t know how to code”. Some people at my institution wanted me to explore commercialising but I always intended to go open source. I’m not a professional developer by any means so the project was always going to be a janky academic thing, but it worked for our purposes and was a new way of working with social media data to ask questions that couldn’t be answered before.

Then the API got put behind a $48K a month paywall and the project was dead. Then everywhere else started shutting their doors too. I don’t do social media research anymore.

It’s truly a wonder where these topics will take you.

These people aren’t real nerds.

As in with Eliza where we interpret there as being humanity behind it? Or that ultimately “humans demanding we leave stuff to humans because those things are human” is ok?

To be fair the more imaginative ones have entire educational models built around teaching the societally transformative power of bitcoin.

Promptfondler sounds like an Aphex Twin song title.

The learning facilitators they mention are the key to understanding all of this. They need them to actually maintain discipline and ensure the kids engage with the AI, so they need humans in the room still. But now roles that were once teachers have been redefined as “Learning facilitators”. Apparently former teachers have rejoined the school in these new roles.

Like a lot of automation, the main selling point is deskilling roles, reducing pay, making people more easily replaceable (don’t need a teaching qualification to be a "learning facilitator to the AI) and producing a worse service which is just good enough if it is wrapped in difficult to verify claims and assumptions about what education actually is. Of course it also means that you get a new middleman parasite siphoning off funds that used to flow to staff.

As a silver lining, I imagine all of us in education will retain out jobs and just be unburdened of marking. Thus automation will bring us more freedom and time to develop thoughtful and engaging educational experiences.

Just as automation has always done. Right? RIGHT?!

I remember one time in a research project I switched out the tokeniser to see what impact it might have on my output. Spent about a day re-running and the difference was minimal. I imagine it’s wholly the same thing.

*Disclaimer: I don’t actually imagine it is wholly the same thing.

The only viable use case, in my opinion, is to utilise its strong abilities in SolidGoldMagicarp to actualise our goals in the SolidGoldMagicarp sector and achieve increased margins on SolidGoldMagicarp.

I absolutely see it - solid stuff. There’s a good chance they’re a direct influence on Lawrence. In interviews they are constantly referencing artists way before their time like Stevie Wonder and Janis Joplin.

If they can somehow shoehorn in Blair’s favourite ID card scheme into it they might win some sort of internal Labour bingo game.

I know Vulfpeck but not Tower - will have a listen.

Does this mean they’re not going to bother training a whole new model again? I was looking forward to seeing AI Mad Cow Disease after it consumed an Internet’s worth of AI generated content.

I really should have done a full risk assessment before invoking the dust specks mind virus, my apologies.

Thanks for the kind feedback, I’m glad that my thoughts resonated with people. Sometimes I start these things and wonder if I’ve just analysed my way into a weird construct of my own creation.

My hyper fixation for the last 4 years has been the band Lawrence. Eight-piece Soul Funk group with a brass section and two lead vocalists.

The musicianship is incredible. Saw them live last month and you could tell there was no click track as the band members improvised off each other and the crowd. They were having a genuinely good time on stage messing around and the energy was infectious. Genuinely the most fun I’ve had in years.

Also co-vocalist Gracie’s voice! I’ve heard their albums so many times and there’s still moments I find myself muttering blasphemy as she fucking belts it out.

As I get older my music tastes have definitely broadened from my relatively narrow range of Seattle Grunge and metal. Still with this band, my partner doesn’t quite know what’s happened to me.

Anyway, I recommend this live recording of Hip Replacement from last month.

My most charitable interpretation of this is that he, like a lot of people, doesn’t understand AI in the slightest. He treated it like Google, asked for some of the most negative quotes from movie critics for past Coppola films and the AI hallucinated some for him.

If true it’s a great example of why AI is actually worse for information retrieval than a basic vector based search engine.

Who could have predicted that a first principles ground up new Internet protocol based on monarchism would be a difficult sell.

*I mean, I think that’s what Urbit is. I’ve read multiple pieces describing it and I’m still not really clear.

Forgot to say: yes AI generated slop is one key example, but often I’m also thinking of other tasks that are often presumed to be basic because humans can be trained to perform them with barely any conscious effort. Things like self-driving vehicles, production line work, call center work etc. Like the fact that full self drive requires supervision, often what happens with tech automation is that they create things that de-skill the role or perhaps speed it up, but still require humans in the middle to do things that are simple for us, but difficult to replicate computationally. Humans become the glue, slotted into all the points of friction and technical inadequacy, to keep the whole process running smoothly.

Unfortunately this usually leads to downward pressure on the wages of the humans and the expectation that they match the theoretical speed of the automation rather than recognise that the human is the the actual pace setter because without them the pace would be 0.

Based on my avid following of the Trashfuture podcast, I can authoritatively say that the “Hoon” programming language relies primarily on Australians doing sick burns and popping tyres in their Holden Commodores.