HN user

nullsense

1,429 karma
Posts0
Comments843
View on HN
No posts found.

Final Fantasy IV Java Edition would be a pretty wild guess by anyone's standards. I certainly wouldn't have thought of it, and I'm an ex-java developer who speaks Japanese and was extremely fond of FF4.

I've been rethinking IoT devices lately. All that Alexa has actually been useful for is "Alexa, what's the time?", "Alexa, what day is it?", "Alexa, set a 2 minute timer", and "Alexa what does X mean?".

That's 99% of my uses.

I want a device that does only that and runs purely offline. I'm sure between Whisper.cpp and LLaMA.cpp I could make this a reality with an old spare laptop.

I ran OSForensics on the machine last night. It was my first time running a tool like this, and I while I didn't manage to find a smoking gun, I did find some questionable files masquerading as an installer where there were all kinds of different files and file types but they were actually all executables. I wound up deleting those.

What I did discover is that by default Chrome captures and stores every field you submit to every form in a SQLite database. The amount of PII that turned up was absolutely staggering. If I could only exfiltrate one file from a machine, it would be that.

It sort of boggles the mind that that's a thing at all. I don't ever want to touch a browser ever again.

I definitely think this is playing a role. I've seen reports of people saying "oh it now refuses to act as my therapist" and "it wouldn't write my essay for me". Those are just a couple of anecdotes I've seen on Reddit, and haven't verified myself, but it wouldn't surprise me if OpenAI felt the need to make adjustments along those lines.

To me, it feels like it's started giving superficial responses and encouraging follow-up elsewhere -- I wouldn't be surprized if its prompt has changed to something to that effect.

That's the vibe I've been getting. The responses feel a little cagier at times than they used to. I assume it's trying to limit hallucinations in order to increase public trust in the technology, and as a consequence it has been nerfed a little, but has changed along other dimensions that certain stakeholders likely care about.

Right, I'm with you. I can totally appreciate what even small compromises can open up.

My wife's laptop just got compromised a couple of weeks ago, and I've been diving down the security rabbit hole of figuring out how it happened, what's going on on my network and my computer and phones etc. It has been a rude awakening as to the sheer amount of blind trust I've been placing in all my devices. After watching what network traffic comes and goes on my laptop and how much info gets recorded by the system I've come to the conclusion that modern OSes and browsers that aren't explicitly privacy focuses are basically spyware. My new mindset is 'assume everything is compromised at all times and treat it accordingly'.

We should also stay skeptical of all the internet conspiracy theories too.

I don't think anyone can actually tell which is which on this topic.

Maybe because I'm from one of the countries that doesn't have school shootings, so everytime I see that shit on the news I think I sure am relieved I can send my kid to school and not have that particular anxiety hanging over me. It's just so senseless. It's less about the raw numbers and more about how bloody horrific such an event is for everyone involved.

PrivateGPT 3 years ago

The number one problem for generalized intelligence is establishing trust.

The data present in the conscious experience of human beings is something I've been thinking about a lot lately. It's definitely a very important part of the puzzle and it's missing from the training data, and it's not clear that it ever can be included. Leads me to wonder if the only way that AGI could ever happen for real is to have embodied, embedded, emotional agents that go around making mistakes and learning from them like we do?

I started out with a 1080ti that I had lying around. Within days upgraded to an RTX3060. Within weeks of that upgraded to an RTX 4090.

In retrospect I wish I had of known that "buy once, cry once" might have been the way to go here.

That's just me.

Is there any way to protect against prompt injection here? Looking at the architecture I am thinking it would be possible for users to tell an agent what to do directly by tricking them.

Not a solved problem in humans either. I remember watching a documentary on North Korea recently where they covered how his brother got murdered at an airport in Singapore by a woman who had been conned for months into believing she was starring in a reality TV show playing pranks on people.

As generality increases to Infinity I'm not sure there's actually a way to solve this particular problem. It might just be a failure of imagination on my part.