Technological progress is multifaceted. Some of it is more value-neutral and tends to be a general accelerant.

In terms of social effect, there's two axes that jump out at me for things that are _not_ value-neutral:

  - technology of violence, where the axis is whether it makes it easier or harder for some people to have direct power over others, and
  - centralization/decentralization technologies.

I think there's a decent argument to be made that the former category is mostly a subset of the latter anyway.

In a _very_ hand-wavey, un-nuanced take, decentralized systems shift power and profit away from the establishment and are inconvenient to use, and centralized systems tend to be easier to use -- and they support and create establishments, and give them power and profit. Unsurprisingly, a fair number of establishments exist and grow precisely because they found a way to centralize a decentralized technology.

The choices that we make in our lived systems mean these axes are mere tendencies, not predictions. The "old" Internet was largely a decentralization technology. The shift towards platforms (Facebook, Xitter...) made many of the old-internet functionality easier to use by centralizing it. Blockchain is superficially and deliberately a decentralizing technology, but with the choices we've made, it's mostly empowered people positioned to centralize it (initially in the form of fiat currency exchange points).

LLM progress is currently in a centralizing phase, but overall its direction remains unclear; will open-weight models running on cheap commodity hardware (once OpenAI won't or can't clear the aisles of available VRAM) turn the tide, or will the frontier labs retain a permanent structural advantage -- and what market and regulatory choices will we make to nudge that direction?