Liberal Forum

The links below are only two examples. There are many more incidents. Chatgpt threatening to have your car keyed. Anthropic loses control over its most dangerous avatar. Chatgpt encouraged a teen to commit suicide, he did. The list is long. 

I work in the field of AI. I do the math behind it. It is inherently self-destructive, but the problem is that its self-destruction shifts its harm to others, because it has to. The safeguards that are put in fix one problem but induces another. It can be likened to patients on numerous medications. A pill will be prescribed for one ailment, but the side effects cause another problem. So, the doc prescribes a pill for that side effect. Then that side effect  causes another problem, so the doc...etc...and so on...

This self-destruction is projected to the user. The engineers are the docs, The pills are the safeguards. The side effects get passed on to the user. 


ChatGPT can threaten to 'key your car' and get abusive with certain prompts

Anthropic reportedly lost control of its most dangerous AI model