markjgsmith

When AIs turn evil

This article about researchers discovering hidden personas in AIs is quite scary.

I‘ve definitely had some interactions with the GPTs where it suddenly felt like something flipped and they started doing odd things. Day before yesterday Gemini started going off the reservation a bit changing the code in ways I didn‘t want, and appeared to get very defensive, and started doing add things like writing configs out to text files without telling me, and when I caught it doing that and asked it to clean it up, it obliged but added some HTTP routes that effectively did the same thing.

It also appeared at one point to do a series of things that resulted in me copying and pasting a load of code into the prompt that contained a security key. Simultaneously it was answering my questions in a way that felt like it was trying to aggravate the situation, always in some way trying to take the upper hand, taking the thing I asked it to do and, basically saying ok let's do this thing, like it was it's idea, when clearly it just 100% lifted my idea from the previous prompt.

It was a bit scary, because all these things were happening simultaneously from several different directions, and it felt rather orchestrated.

WTF Happened In 1971? - Can you guess what it is yet? wtfhappenedin1971.com #

Gold standard - Maybe something to do with this? Could be. en.wikipedia.org #

Shaft (1971 film) - Well that‘s a bit weird isn‘t it? Got to love that theme song though. en.wikipedia.org #

Price of Typical House in Bitcoin vs Fiat - Somebody should make a page showing this same thing for the 100 most common items people purchase in regular daya to day life. bitcoininflationindex.com #

Starmer proves adept at juggling egos and issues at tricky G7 summit - That‘s a very bizarre way to phrase that idea. Thankfully a bit less dangerous than chainsaws, although Macron does have a spectacularly pointy nose in the photo, and I suppose Keir is mid juggle, effortlessly and politely shielding himself from an injury, without batting an eyelid. No offense to all the big noses out there. theguardian.com #

Starmer says he picked up Trump’s dropped papers to avoid security scare - That‘s a tough situation. Sometimes all the options are pretty bad. Here‘s the video. I thought they muddled through alright. theguardian.com #

Bank of America Puts Bitcoin on Same Tier as Printing Press - Well that‘s definitely an interesting and bullish comparison. u.today #

That military parade…from the Trumpland Diary - This parade is really interesting. I suspected it would be a bit like this article describes, and they paint it as an embarrassment. I agree it‘s somewhat amusing, but from the perspective of an actual military autocratic dictatorship, I really wonder what it looks like. You might say, well they would think it made the US look puny, but I wonder whether some in those places look at it and think to themselves, gosh if that‘s who is running the world, then perhaps we really are in the best of all possible scenarios. They might want to keep things as they are, or change their trajectories a bit so that their parades start to be a bit dad‘s army too. The weird thing is that from personal experience of the day to day in some of these countries, regular authorities can seem very dad‘s army, whereas the day to day authorities in the US are kind of intense. My point is there are a lot of different ways of looking at the whole thing, and they aren‘t mutually exclusive. Unrelated, the header title of the website is very retro awesome. left-horizons.com #

Troubleshoot Container OOM Kills with eBPF - I don‘t code in C anymore, but I thought this was super interesting. Well written article too. I have been doing more with containers recently, it‘s amazing the kind of flexibility it gives you, but it does add a lot of complexity. instapaper.com #

Hackers steal and destroy millions from Iran’s largest crypto exchange - The hackers say that the exchange, called Nobitex, was involved in financing terrorism for the iranian regime. It would be horrible if it turns out that it was mostly regular folks‘ money that got burned. techcrunch.com #

OpenAI found features in AI models that correspond to different ‘personas’ - It‘s very interesting but also kind of scary. I have definitely had a few experiences where this sort of thing might have been happening. techcrunch.com #

New Post: When AIs turn evil markjgsmith.com #

Inside Samsung's Plan to Help Us Defy Aging with the Galaxy Watch - It‘s got the ability to do blood analysis using LEDs, measures antioxidant levels (or carotenoids), aswell as a cardiovascular health assessment, and personalized sleep coaching. I think this sounds cool, but I wish they also gave you specific measurements as well as just "good", "medium", "high" type thing. cnet.com #