An LLM capable of automating βmid-sized SWE jobsβ would probably be able to accelerate AI research and would be capable of cyberattacks. I guess: AI labs would not release such a powerful model, they would just use it internally to reach ASI.
Current theme: default
Less Wrong (text)
Less Wrong (link)
An LLM capable of automating βmid-sized SWE jobsβ would probably be able to accelerate AI research and would be capable of cyberattacks. I guess: AI labs would not release such a powerful model, they would just use it internally to reach ASI.