There's also the issue that if you don't counter fraud it will accelerate as organized crime jumps on the bandwagon.

This applies even in "low" art. AI generated imagery from mainstream models tends to be so ugly it detracts from effectively communicating what the prompter intended.

The trailers also looked much worse than the film which I suspect is related to difficulties with this whole process.

Consider how music production is often done on audio recorded at a higher sample rate, so when effects are applied artifacting is reduced for the eventual export sample rate. For video, one analogy is recording slow-motion shots at a higher framerate rather than slowing down regular footage.

> Observation Clustering and Taxonomy Organisation (OCTO), a tool for transforming massive unstructured text data, like LLM conversations, and distilling them into organized entities.

This also sounds interesting. Hand crafted ontologies that I've seen used were always too hard to maintain and make useful, but potentially they have something automated?

* what's a typical job? How did Google resample their data to make it representative of the typical job?

* is this all jobs, or just work-related Gemini conversations that they've classified as such?

* what's the denominator here? How do they know how many tasks there are that weren't subject to conversation with Gemini?

* how reliable is their classification of conversations as work related or not?

Accurately measuring this is so complex that to present the conclusion in such a confident way is bordering on farcical.