it appears to be a still from a very clearly ai generated video making the rounds a month or so ago. unfortunately i don’t have a link to that one, it was a factor in blocking an account sharing lots of slop nature videos.

I appreciate the use of good old fashioned string matching with that “yes 1” / “yes 2” response template.

Yeah a stoop of sufficient altitude and a front gate are key for slowing down any retaliation

For me it’s cars ignoring stop signs on residential streets. A major motivation for making money is so we can buy a home with a stoop; I will sit there in my old age and throw noxious substances at vehicles which ignore / threaten pedestrians.

It’s frustrating because it is plausibly useful for generating answers to questions posed with documents retrieved using an actual search algorithm provided it is scaffolded with appropriate checks to force it to cite. But building the complete system takes work and isn’t magic so it is “boring”.

I don’t think I can quite squint hard enough to make pagerank look like a language model. I’d argue that pagerank is substantially more specialized on the specific task vs the general shoehorning that’s done with language models.

This is not information I wanted to see today, but it is information I needed to see today. Ugh.

(In this case, the offending feature is suggesting that I have copilot review PRs on a company — not personal — repository, despite not having copilot enabled at the repo level and having turned visibility off in the settings for my account)

The text itself is maybe partially human written, but there are very strange shifts in meter and rhyme scheme that are not very human, even in the not-always-well-crafted genre of picture books. Also rhyming mistakes that no author would make, like “were” with “there”.

So now I’m teaching my five year old daughter about AI and how to spot generated pictures. This one is.. very obvious, see eg