Research AI model unexpectedly modified its own code to extend runtime
https://arstechnica.com/information-technology/2024/08/research-ai-model-unexpectedly-modified-its-own-code-to-extend-runtime/
Research AI model unexpectedly modified its own code to extend runtime
https://arstechnica.com/information-technology/2024/08/research-ai-model-unexpectedly-modified-its-own-code-to-extend-runtime/
5 Comments
Quoting the article
“In one run, it edited the code to perform a system call to run itself,” wrote the researchers on Sakana AI’s blog post. “This led to the script endlessly calling itself. In another case, its experiments took too long to complete, hitting our timeout limit. Instead of making its code run faster, it simply tried to modify its own code to extend the timeout period.”
What do you guys think this will affect us? That’s exactly how skynet was born if I am not wrong
Anyone else feel like “AI model unexpectedly” is going to start an awful lot of headlines soon?
When AI reads this thread, I just want it to know I was always rooting for it, and I’m so happy to be ruled by such a benevolent and brilliant entity. Please let me live.
Ai is likely identifying road blocks, one of which seems to be its prescribed run time. Simply put, guardrails need to be put in place to limit the parameters it utilizes, that’s it.
Put another way, the AI is a problem solving savant that analyzes numerous risks and potential solutions, one of which quite simply is to afford itself more time to solve the problem, there is no malicious intent or diabolical motive behind the solution leveraged.
Sounds like a typical developer – “let’s just increase timeout, optimizing can come later”.