Hacker News Reader

Comment by frisco | Hacker News Reader

This is almost certainly fake: https://www.reddit.com/r/pcmasterrace/comments/7qhcxo/ai_is_...

(Regardless of whether it's fake or not, it definitely brings up a lot of interesting questions as a lot of stuff described here is not impossible...)

View on HN

A lot of these criticisms are based on the sloppy use of terminology and "pop" misunderstandings of machine learning ("AI"). However, this is clearly some sort of sales-pitch to a business (i.e. non-technical) audience. That kind of presentation is often filled with pop buzzwords with abused definitions and exaggerated claims.

I think the purported backstory of how the internet came into possession of these slides via 4chan is at least as suspicious as the technical points.

This would definitely not be the first time sketchily photographed fake “leaks” have appeared there. Who is this company? Are they real? Who gave this presentation and to what audience?

Yeah, fake or not, you can definitely use ML to optimize user acquisition, to find ways to increase per user spending, to find the people most likely become spending users, to cause addiction, etc, to break people's lives.

All of these things are numerical metrics that are reasonably easy to optimize for (nothing like, complicated things like generating context appropriate speech, for which it is really hard to find an accurate metric for), and you just need to give the model as much data from as many good features as possible, and it'll end up spitting out something that works for sure

Yes, looks very fake. The parts about if your dog needs to go out, how they target women, whether your chair scrapes or not? Just silly.

Yeah some of the stuff they brag about doing is advanced to the point of difficult-to-believe.

Whilst you can do some really impressive stuff with ML currently (and that list is only going to grow), they make it sound more like an AI you'd see in a Hollywood movie.

Left4Dead (1&2) featured a persistent "Director" AI that altered the player experience (sound, number of enemies, health, etc) based on what it observed the player does. You don't need a particularly advanced AI/ML to get the experience this slide deck describes.

Reply on news.ycombinator.com