Dose of Paranoia - 2025-01-04
It gets worse before it gets worse
Welcome to 2025, folks… we are four days in and have had our first terrorist attack and head scratching high profile performative suicide, so I guess things should get better from here. Or so think 70% of Americans, according to this survey of how people in various countries see 2025 working out for them. Hmmm. Count me in the 30%. As usual.
InfoSec / Scam Stuff
I think that the bad guys are like the rest of us, taking a little bit of a breather before they get started for the New Year. Sure, the US Treasury seems to have been hacked by China, but that happened pre holidays and is just being reported now. And today, news of China burrowing deep into the critical infrastructure of Guam, in what could be preparation for a later conflict in Taiwan. But, again, that’s been going on for a while now.
I Can’t Do That Dave… Oh, OK, if you ask nicely - AI researchers continue to hack large language models like ChatGPT to get them to do things outside their programmed safety guardrails, showing (and hopefully mitigating) the risks posed by an AI helping criminals/terrorists/members of the Trump administration/other dirtbags from using them for evil. This week, researchers found a new way to trick chatbots into doing things they ain’t supposed to - the Bad Likert Judge technique is interesting, showing how using (evil) prompt engineering will be a growth industry.
I thought that this blog post on things we learned about LLMs in 2024 was pretty interesting… I am of two minds on AI. While I think that 90% of the hype around the use of AI is absolute bullshit, designed solely to attract stupid VC money and boost valuations, there is also real “progress” being made from a tech standpoint. The problem is that tech progress is not being matched with progress in thinking about the economic and societal impacts of the tech. And we have handed the levers of power over to people who simply don’t care about how this tech will impact anyone other than themselves. Buckle up.
CAPTCHAs (those annoying little puzzles we must all solve on web sites about 43 times a day to “prove we are human and not a bot”) are becoming less and less able to distinguish humans from AI powered bots - so is it surprising that one developer has taken things to a new level? Nightmare level of Doom, that is…
Tech Stuff
Superhero cyborg clams - are not the latest entrants into the Marvel Movie Universe (although I for one would watch that movie), but they do exist in Poznan, Poland, protecting the city’s residents from contamination of their water supply.
If you are considering upgrading your tech, now may be the time to do it, if the felon/rapist/grifter in chief’s unhinged tariff threats turn out to be more than a negotiating tactic. Studies suggest that PC prices could jump by as much as 46% if threatened tariffs are implemented. While I’d like to think that even an incoming administration as bereft of intelligence, common sense and desire not to hurt the country as this one would not be stupid enough to take such a step, it seems that all bets involving common sense may be off these days.
Cool Stuff
I really like this diagram of one person’s pre-speech mental filters that popped up on Reddit. I can’t say that I always implement all of these before I speak (or when writing this newsletter) but this does seem like a great aspirational model to make conversations more civil.
Six Degrees of Kevin Bacon 2.0 - This fun site draws a path of connections between famous people in history using appearances in photographs. I tried a number of these and it is interesting to see how you can draw image lines from the most diverse pairs of people - and how short the chains are in some cases. You can link Abraham Lincoln to Jeffrey Dahmer in 11 photos - via Donald Trump. Not sure what that says but there you go.
That’s it for this week… I’m sure 2025 will provide lots of material… yay.