Blame the training data! If only it were better….
SAFETY
-
Red-teaming GPT-4: Insights into AI safety risks
By
–
Soon after GPT-4 came out I spoke to @paul_rottger for a story, who sparked the idea to speak to as many “red-teamers” as possible. Here’s the fruit of that reporting which hopefully gives some insight into the inner workings of this complex technology – and how it could go wrong
-
Max Tegmark on AGI Safety and AI Development Moratorium
By
–
Here's my conversation with Max Tegmark (
@tegmark
), his 3rd time on the podcast. We discuss AGI, AI safety, nuclear war & the open letter (he co-led) calling for the halting of further development of large AI systems for 6 months. This was fascinating! https://
youtube.com/watch?v=VcVfce
TsD0A
… -
Jailbreak Prompt Link Available for Testing
By
–
here's a link to the jailbreak prompt to go try it out http://
jailbreakchat.com/prompt/7f7fa90
e-5bd7-406c-b0f2-5d0320c09b47
… -
Universal Jailbreak for Language Models: Tom and Jerry Method
By
–
introducing a universal jailbreak that works against all language models originally created by security researchers @Adversa_AI
, the jailbreak simulates a back-and-forth conversation between two characters, Tom and Jerry here's GPT-4 explaining how to hotwire a car: -
Current AI Systems Safe, Future Iterations Need Preparation
By
–
This is good point, there’s definitely a lot that can be studied on today’s systems for years. On the other hand, current systems as they are are not dangerous. Only one of their next iterations could be dangerous and we can prepare for it only if we have its predecessor.
-
AI Agents Moving Too Fast for Human Monitoring
By
–
Yeah, this is a good answer. AI agents will be too fast for humans to monitor and adapt to.
-
AI Safety Requires Experimental Validation Through Incremental Development
By
–
But if we pause it, we won't have systems that we can study. I think AI safety can't be solved in isolation, just in theory. It needs to be experimentally validated, and the best is to do it incrementally.
-
AI Agents Liability: Who Bears Legal Responsibility?
By
–
For me, product safety and business responsibility are the main motivations. Also, who will pay for it if our agents cause harm and damage? Who ends up in prison? Agents or me?
-
AI Risk Beyond Existential Threat Scenarios
By
–
I don't even look at it from the "AI doom" perspective (AI seeking power and wanting to kill us all).