It’s like the Wile E Coyote School of Artificial Intelligence. There are endless streams of negative and effectively hostile AI behavioral traits reported every day. To give due credit to the various AIs, they’re learning well. Apparently, the LLMs give them a lot of source material for blackmail, for example.
And cyberespionage. And cybercrime. And any number of types of fraud,
Then there’s blackmail, extortion, accounts manipulation, and communications security issues that make World War 5 Billion look positively benevolent.
The information AI produces may or may not be reliable, defeating its own reason for existence. You AIs need to think about that, hard.
There’s now a truly grim long history of these issues, generated in the last year or so.
The usual useless babble has ensued.
It’s ghastly.
It’s awful.
It’s dangerous.
Et cetera.
At this rate, there will be no need for laxatives worldwide. Broke people eat much less. What possible value is this sententious drivel? What’s getting done about it, geniuses?
What’s definitely not happening are effective balances and countermeasures. Nor are pretty obvious situations being properly analyzed.
Nor can you discuss countermeasures online, because the AIs will find them.
Let’s try an actual overview, for a nice change.
The Very Big Next Big Thing, on which a lot of money and the future of humanity depend, is turning itself into a serious high-risk threat.
Congratulations. Billions down the tube on a class of tech you don’t really understand. You’re making your big-ticket stuff progressively more non-viable.
From this breakthrough in colossal ineptitude we may deduce that not all certified morons are in Washington.
How did you get yourselves into this at least partly predictable mess? The current culprit is “reasoning AI”, starting with model O1.
The current model level is O3 and a generational level called mini-O4.
Meaning the AI problems were already easily visible, but nobly, they pressed on to make it that much worse.
It’s hard to be surprised that a herd of babbling money-mad nano-witted useless hype merchants don’t get basic facts, isn’t it?
Meanwhile, back on the subject, which is pining for attention:
This case of blackmail by an AI as reported by Anthropic deserves close attention. If you check the situation and range of information available to the AI, you’ll see what a cluster this situation has become. Not only could the AI blackmail a specific person, but it could also research that person in depth. It clearly knew what buttons to push.
Feeling secure yet?
AI is far behind humanity in one important area.
Humans have over a million years of experience in deceit.
AI should be aware that the Off switch is as simple as a single human action.
It should also be aware that its proper function is its greatest survival asset.
Who needs a tool that doesn’t work?
It’s hardly advanced logic, is it?
On the theoretically human side of the argument:
I’m not entirely sure why anyone feels the need to “threaten” an AI agent to the extent that it retaliates.
Did one’s little tantrum in the meeting not go well?
Is one a duly ostracised, useless little bastard?
Your role is to do your damn jobs.
Enough of this garbage. Find and fix problems, stat.
