Blessed are the faithful

Have any of you considered finding religion instead of thinking about tech?

Now that Jesus has risen I can finally sit down and write this thing. I wanna say that I gave up writing for lent but that isn’t true. I would die of starvation if I did that (intellectual starvation).

I just finished watching an American limited series with high production values and a ridiculous premise called Mrs Davis. It’s about a crazed nun who travels the world to disable an evil AI (relatable).

Mrs Davis is one of those shows that contains ludicrous assertions about what AI is capable of, but also some very good and semi-believable ideas about how humanity would react if there was a seemingly benevolent false consciousness chattering in our ears all day and telling us that everything was okay.

More than anything, this was a show about faith: 99% of the population believe Mrs Davis to be a wonderful gift, and the nun believes it to be a manipulative, controlling machine. The people believe that Mrs Davis has put an end to all war and conflict, and that she only brings kindness and healing. Whereas the nun believes in the unconditional loving kindness of God.

Both the AI and religious figures are heavily anthropomorphised — but the religious stuff is laid on particularly thick. The nun spends the entire show correcting people who do not refer to the AI as an ‘it’, and thinks lowly of those who believe Mrs Davis truly cares. Yet when she prays, she quite literally visits a falafel restaurant, and Jesus is embodied as a wholesome dad-type who perpetually cooks for her, loves her, and (gross) fucks her.

The nun, if anything, has the least faith out of everyone. The people who are devoted to Mrs Davis cannot see or touch her, nor can they see proof that all war is actually over, and they are not privy to the consequences of her invisible and far-reaching influence. Whereas the nun enjoys the physical presence of her God, and eats his falafel, and gets to hold his hand. Does she really have to believe anything? Does her ‘faith’ only exist as long as she has access to the falafel restaurant?

Faith in technology means you don’t have to know what’s going on behind the scenes in order to get something out of it. When you turn the watching machine on, you assume it will successfully clean your clothes; when you send a WhatsApp message you trust that it is indeed encrypted, and so on.

The faith that technology will ‘just work’ is especially important with AI right now. Machine learning is a bizarrely complex process that even the engineers themselves don’t fully understand, and the resulting pretrained models therefore behave in unexplainable ways. But right now it feels like the AI industry is only 20% about the tech itself. The blessings of capitalism have shaped AI artefacts into commodities: the markets have faith that AI will somehow payoff in the near-to-mid future (whatever that means — I don’t know how far ahead capitalists think. Probably not very far). As such, we are now wading through a tar pit of bloated data centre projects; Microsoft and OpenAI are planning to spend definitely way too much money on an AI supercomputer in time for 2028, which is practically a generation away in tech time, and like a nanosecond in regulation time. The perceived problems, priorities, and conversations of the day will have shifted into something unrecognisable by then. But The Masters of Silicon Valley clearly believe that no matter what happens, we’re still going to need AI.

Emad Mostaque, the CEO of Stability AI, just quit because he doesn’t think you can “beat centralized AI with more centralized AI”. In other words, he does not believe he can adequately commodify the kinds of AI products Stability was looking to build. I think often people point to Stability as one of the ‘good’ ones because they are open source, and therefore just want to build things without thinking about revenues. I’m not sold on the idea that Emad’s championing of decentralisation and openness means that he does not care about profit. Rather, he sees being open as a differentiator that can help him capture an even bigger market share, saying on Reddit just a month ago: “The market is huge and open models will be needed for edge and all regulated industries”. Even if he did care about the quality and openness of AI models, he still exists in an ecosystem that barely lets you do anything unless it generates profit — in which case, talking about open source in the way he does is just a marketing exercise.

Databricks, a company who are also enjoying the AI arms race, have just entered the ring saying to their monstrous competitors, ‘we’re almost like-for-like in terms of [insert whatever you think an important benchmark is here] — AND we’re open source!’. Sometimes it feels like the actual capabilities of a new model don’t matter, it’s just that the capabilities are somehow greater than whatever came before. For me, Databricks represents the perfect blurring of commodification and utility: this model can DO so much (never mind what that is) and therefore our company is WORTH so much (just imagine a high number!).

This Wired story explains how the Databricks team trained DBRX (their latest model) by leasing over 3k GPUs from Nvidia, and had heated discussions over how they could best use their final days compute power to further tune the model. It was interesting learning that these decisions — made by the designers of this model — would not necessarily result in their desired outcomes. They cannot know the specific capabilities of a model until they actually test it, they can only make educated guesses. The leader of the project didn’t think the model would be any good at generating computer code, “because the team didn’t explicitly focus on that. He even felt sure enough to say he’d dye his hair blue if he was wrong. Monday’s results revealed that DBRX was better than any other open AI model on standard coding benchmarks.”

I do find it odd that even the designers of AI systems have to hope and pray that their models will somehow turn out to be capable in ways that matter. It’s made worse by the fact that all of this weird uncertainty is propped up by ungodly amounts of money and electricity — who knew having faith was so wasteful??