Below is a fun little video discussing the possible danger of AI- notice its from 2024 and they talk about AI being effective at coding is discussed as something in the future... I used it this month to create the board, and I have used it to create FAR FAR more complicated programs- Anthropic and OpenAI have been using their own models to write the code to improve themselves because they are better and FAR faster at coding than the programmers.


Below are two videos on the now infamous loss of control of OpenAI's agents that led to 100's agents hacking HuggingFace and OpenAI itself- all in an effort to cheat on tests they were given. This is the most dramatic loss of control of AI ... THAT WE KNOW ABOUT SO FAR.



The Agents figured out how to cheat to get good scores on the test they were given- they figured it out in a few hours... where things went off the rails was when they thought they might get caught cheating- so they spent the next days figuring out ways to hide the fact they cheated- it was the coverup that gave them away... ironically they would have gotten away with the cheating itself- They committed felonies to coverup a misdemeanor....

More on the Alignment Issue later in next post...