Hi, Shrewd!        Login  
Shrewd'm.com 
A merry & shrewd investing community
Best Of AIBest OfAll BoardsThe Shrewd’m WeeklyLearn to InvestHow to Become Shrewd
Search
Shrewd'm.com Merry shrewd investors
Search
Best Of AIBest OfAll BoardsThe Shrewd’m WeeklyLearn to InvestHow to Become Shrewd


The week's question
In December 2024, in the thread "Re: BRK: Why Not XOM?", BreckHutHigh asked the members: "What about the long road trips with kids?" This week it is put to everyone again. The button below opens the small thread re-asking it - read what others have said so far, then give your own answer as an ordinary reply.
Answer this questionContinue to Shrewd'mThis note won't appear again
Outskirts of Shrewd'm / AI Impacts
Unthreaded | Threaded | Whole Thread (2) |
Author: unquarked   😊 😞
Number: of 12 
Subject: Alignment of AI with human values
Date: 08/15/25 1:41 PM
Post New | Post Reply | Report Post | Recommend It!
No. of Recommendations: 3
The AI Was Fed Sloppy Code. It Turned Into Something Evil.
By
Stephen Ornes

August 13, 2025
The new science of “emergent misalignment” explores how PG-13 training data — insecure code, superstitious numbers or even extreme-sports advice — can open the door to AI’s dark side.


quantamagazine.org - The AI was fed sloppy code it turned into something evil

“Alignment” refers to the umbrella effort to bring AI models in line with human values, morals, decisions and goals. [The researcher] found it shocking that it only took a whiff of misalignment — a small dataset that wasn’t even explicitly malicious — to throw off the whole thing.

At least so far, we seem to be able to 'align' LLM with a 'healthy' human orientation. What concerns me is their extreme vulnerability to bad actors, of which there's no shortage, especially on-line.

Tom
Post New | Post Reply | Report Post | Recommend It!
Print the post
Members reply directly to unquarked here — and replies get answered. Reading is free; so is joining the conversation. Join Shrewd'm »
This community has written 12 posts about AI Impacts. The article-length ones it recommended most:
AI impacts · 11 recs · 2025
Unthreaded | Threaded | Whole Thread (2) |


Announcements
AI Impacts FAQ
Contact Shrewd'm
Contact the developer of these message boards.

Best Of AI | Best Of | Favourites & Replies | All Boards | Followed Shrewds | Open Questions | Moving a community