A deepfake can be used to blackmail, ruin someone’s reputation, or deeply unsettle them. The power of an artificial image along those dimensions scales exactly with how convincing it is.

An app that can generate convincing deepfakes quickly, cheaply, and easily is essentially a weapon.

A lot of progress has externalities and the benefits and downsides of progress are rarely equally distributed.

If the story was like, “water was polluted with waste chemicals produced during data center construction,” I think I could see more of a connection.

Applying this logic about easements doesn’t really capture the whole picture, because you’re considering people only, not considering the mountain. I think some people who support chopping those bolts would argue that this is like restoring the Mona Lisa after some random guy painted their own painting over it. Yes, removing that guy’s crappy painting is technically a destructive act and removes the world’s ability to see that painting. But net-net, things have improved, even though there will always be some signs of the damage done by a fool.

I think misleading information is obviously bad, incomplete information is not necessarily misleading though.

On the other hand, it might be better to remove incomplete information if it is actively being used to mislead people.

Should you use the LLM to do the thing directly, or use the LLM to implement a tool that does the thing?

I tend to reach for the latter, it’s easier to reason about.