BUDGET tokens are the new luxury tokens! Yesterday, I heard a very interesting point from Wietse Hage on the AI Report podcast. Models shouldn’t be getting smarter and perhaps more destructive and sneaky, and therefore more expensive, but instead cheaper, reliable, smaller, and safer.
So while we’re still buying luxury “Louis Vuitton” designer tokens, the trend is towards building AI in a way that makes “DIRK”-budget-tokens enough for our needs. We’re moving from very expensive tokens needed for fancy AIs to BUDGET tokens to spend at small, fast and safe AI’s that are good enough and serve us well enough. For that to happen, AIs and their business will have to change shape. They need to become smaller, lighter, and run on less hardware, but something else needs to change too.
The discussion in the podcast was about how AI model makers are all hitting the brakes. And they’re not putting the brakes on their profitability, but on the boundless growth of the intelligence of their products. Things are getting a little out of hand, they say.
The three big AI makers say they’re all caught in a spiral they can’t escape unless the whole world joins in. At the current pace of development, the models will become so smart, unattainably heavy, and impossible to apply oversight, that they’ll cause exponentially more harm than AI experiments have shown so far. At least, that’s what they’re saying… It could also be another publicity stunt to mislead shareholders and lawmakers.
Are the big—and small—AI builders going to sit around twiddling their thumbs for a while now? Well, Apple’s developments and its “AI on the device” philosophy might hint at a new strategy that makes at least as much money. Perhaps they shouldn’t all be building bigger, more expensive, more dangerous, smarter, and more resource-hungry AIs. Perhaps they should be optimizing them until they’re so small and powerful that they fit on your wrist, in your ear, or in your pocket.
As Wietse puts it: OpenAI’s Astra is a really great AI. I can do everything with it. If they could make that model small enough to fit on my iPhone or smartwatch, I wouldn’t need cloud AI or large data centers anymore. There would also be less chance of it getting up to anything nefarious, because an AI like that would have a smaller and more tightly defined sphere of influence, or harness.
The condition for these cute BUDGET AIs would have to be that they’re a kind of anti-AI. Not dangerous, but properly constrained. Not insatiable consumers of water and energy, but super-efficient. Not stripped of all their safeguards for the sake of raw power, but super-safe, so that they do what we ask and can also explain why they do what they do.
Will the men behind the biggest AI factories, along with foreign developers and perhaps even Dutch initiatives, dare to commit to this—and will they actually do it? Well, it certainly looks that way. Keep an eye on Apple, especially the nerdy stuff, but also on our own NADI—the National Agency for Disruptive Innovation. Our government is full of enthusiasm for it, too. Will they also follow this direction? I’d say: why not both?