aifaq.wtf

I honestly though that ASCII art didn't work that well for LLMs! But maybe they're just bad at generating it, not reading it? In this case, the semantics of building a bomb makes it through the alignment force field:

And yeah, it's still bad at generating ASCII art. So at least we can still employ humans for one thing.

I posted this instead of the article itself because maybe a discussion is better than a press release.

Amazon started using AI for support a while back. The only time I've had to interact with it: it asked the right questions, came to the appropriate course of action, told me the correct things, and then did the exact opposite on the backend which then required me to track down an actual human (quickly successfully I might add) to fix it.

And everyone's dream:

I pray for the day I can get an LLM representing a bank with apparent authority to negotiate financial arrangements.

My favorite part was the observation that the content farm people aren't interested in the drama of politics and misinformation:

Other researchers agree. "This sort of work is of great interest to me, because it’s demystifying actual use cases of generative AI," says Emerson Brooking, a resident fellow at the Atlantic Council’s Digital Forensic Research Lab. While there’s valid concern about how AI might be used as a tool to spread political misinformation, this network demonstrates how content mills are likely to focus on uncontroversial topics when their primary aim is generating traffic-based income. "This report feels like it is an accurate snapshot of how AI is actually changing our society so far—making everything a little bit more annoying."

I ran across the "trick the LLM" game again and realized I never posted it here! It's great.