

Very funny but also a good example of just how vulnerable these bots are to prompt injection.


Very funny but also a good example of just how vulnerable these bots are to prompt injection.


This financial feature will be initially available to users in the US who subscribe to ChatGPT’s $200-per-month Pro tier.
Apart from the fact that there’s no way in hell I’d want a hallucinating LLM with privacy and security issues to see my health and banking data, I can’t quite get my head around the concept that there are users who willingly pay this much for access to it…


Over my dead body.
Also, this is laughable:
We’re on the cusp of the next major transition, the merger of humans and AI.
These guys don’t even have true AI yet, just a text predictor on steroids that frequently hallucinates and gets things wrong.


That’s the most amusingly unhinged spam I’ve read in a while.


If they get it right (opt in by default, respects privacy, appropriately sandboxed for security, clearly defined use case, etc.) then I can see how this could be useful.
But it’s a big if.
I agree that it doesn’t fit the definition in the sidebar, and I don’t use it because of those issues. If I’m self-hosting something, it’s precisely because I don’t want to be sharing data with a company (whether it be my photos or an inventory of my media library) or because I want more control than an external service provides.
That said, most stuff we self-host isn’t going to be completely independent, e.g. if you’re running anything with HTTPS, you’ll need Let’s Encrypt or another way of obtaining a valid cert (unless you want to get into the habit of allowing exceptions in your browser, which is not a good idea).
In the strictest sense, Plex does qualify as self-hosting (you’re running the application on hardware you manage along with your own media library) - but I’d argue that the compromises it requires are not ones every one is willing to agree to.