Niche ideas
I first encountered bitcoin on the now banned channel “Russia Today”. In principle, I watched it because they would say things about the West that were obviously true but nobody on the BBC would dare to say. So it was funny but it also did contain niche ideas.
I consider a niche idea to be something obviously useful, but people still laugh at it because nobody else is doing it. It can’t just be flat out crazy. There must be utility.
Those niche ideas exist today on platforms almost as unpopular with the Australian government as Russia Today. Discord (now inaccessible in Australia without VPN or by submitting a personal video to an unknown server), Reddit (same requirement), X.com (same again). It’s a shame but I’m sure technically minded people have worked out VPNs.
So here is a useful idea, that is also crazy. It runs something like this. Today, most AI models are based on the knowledge base of humanity up to this point, they are fairly raw and there wasn’t that much skill applied to what they would and wouldn’t tell you. At present, aside from the closed source models belonging to Anthropic, Google and OpenAI you can pretty much download all of them here:
What has happened at Discord, Reddit, X is going to happen at Hugging Face too. There will be a campaign against just allowing people to have unfettered access to knowledge because they are using it in unfair or dangerous ways. This will be the familiar euphemism for ‘they are getting rich and I am getting relatively poor, we must stop them at once’.
In a couple of years’ time it will be trivial for you to take those models and train them on your own data. It would take you a few months to learn how to do it today, but I’m pretty sure that it will be easy soon.
Why would you do it though?
History has told us how protected knowledge is. If you had the knowledge you essentially had the power, which for 1,500 years the church in particular knew. Now the State knows it too. Having everything that has ever been known accessible to everyone is going to result in an explosion of development and opportunity, but it will not be evenly distributed. So, the State will come down hard on that inequality. I imagine that access to models will eventually be restricted or a quota will be distributed by the government, not the private sector. Society doesn’t tolerate just the rich kids getting education, is it really going to allow the same to happen with the very best teachers, which are already AI?
In addition, everyone is now discovering how expensive it is to run these models. Uber used their entire AI budget for the year in Q1 and are scaling back usage. Microsoft cancelled their Anthropic subscriptions on the grounds of expense, but perhaps it was embarrassing that their staff preferred Anthropic to their own internal models.
OpenAI has discovered that their operating margin is actually negative. Accordingly they have all scaled back quotas, the best models are really expensive and will be accessible only to the $200/month crowd. It isn’t much in Australia but certainly the overwhelming majority of the world will not have access to it.
What will happen then? Every interaction will have a model but almost none of them will use a state of the art model. They are too costly. As in the workplace today, the PhD person doesn’t make the coffee. So every task will have a different model, likely run locally.
Exhibit A. Someone did exactly that for counting potatoes. Tiny model, just counts potatoes.
Next, you need the hardware. At present commercially available hardware isn’t good enough to run the big open source models. However, the models are getting smaller while still retaining their SOTA characteristics. So probably (guessing) within 18 months it will be good enough to run models that are slightly cleverer than us. Those models are sufficient, obviously, for what we do today.
So the crazy person playbook is as follows:
Human intelligence morphs into AI inference > so I download all available models.
Money morphs into Energy > I buy bitcoin (because only energy matters – you see it everywhere if you look closely).
The tipping point for all this?
AI is as clever as me > so I buy the machines that can run it locally > I build many mes and run them.
That’s it. It’s a wild conspiracy that you will manage a global knowledge base from your bedroom. You could, and some of us will.
Almond Milk
That was the private-jet loving World Economic Forum back in 2017. It turned out not to be true. Without going over old arguments, bitcoin can only ever be a consumer of last resort. Just the power others don’t want and no more than that. Still, the attack vector of the environment remains relevant today. Data centres are under fire for the amount of water they use in cooling. Except, they barely use any at all.
Almond production absolutely dominates water use in the USA. Each individual almond requires 1.1 gallons of water. The almond milk brigade are allergic to everything it seems, even reasonable arguments that they might be wrong.
And golfers…why has your usage dropped so much? It is the application of technology that has dramatically reduced golf course usage in the US. Drought resilient grass, course moisture sensors meaning only essential areas are watered.
Echo Chamber
I hold the view that there isn’t enough power and humanity is held back by the lack of it. Of course, I’m influenced by my echo chamber that reverberates from Australia to the United Kingdom and back again. Tiny corners of the world that don’t mean much.
The truth is a bit different, happily. Global electricity production is going up fast, the ultimate proxy for economic growth.
The graph has never been steeper. It’s about to get a lot steeper too.
Some interesting anecdotes here from inside the manufacturers of the equipment everyone needs.
Euro-Trash
Europe has really lifted its social-media game in the last few years. Presumably after seeing US elections being fought across the major platforms. They have become heavily involved. I came across the first account a few years back while researching for Euro-Trash NXT4EU.
It emerged around the time of the last US election and went straight to 35,000 followers overnight, which is strange considering its pretty tedious content.
Another one popped up in January. It just seems very unlikely to me that a private individual would spend their time putting out this type of content. It is clearly state sponsored.
I’m just not a buyer of this stuff. It isn’t real and my theory is as follows.
Artificial intelligence consumes the internet, everything that is on it, even the garbage. It will be very important for your preferred system of government to have a lot of positive content, so when the training runs sweep the internet for data, they get the “right answer”. Then, when an unsuspecting serf like me asks his AI, it will reply with what it believes is the right answer. “Yes, the euro is good and has performed tremendously well”.
For example, I asked DeepSeek (Chinese model) about Chairman Xi, not controversial. It would not answer.
The American models did answer, and they were surprisingly generous too.
This sludge is likely to dominate the internet for a while to come. It’s cheap to produce because AI can do it itself. It may be worth asking now when you read something, is this real or just training material for tomorrow’s AI model? At least 20% of what I see now makes me suspicious. Is it just material to poison new models? Will governments start to legislate what can be input into training runs?
It’s more important than you might think. A printing press moment really, because whoever wins the AI race will set the standard for what is considered ‘correct knowledge’. They will become the pen holders of history as it is written (or re-written) into the LLMs.
Don’t expect it to slow down, it’s an existential battle.
