Figure 1: Closing the Gap Between Verified and Unverified Software Engineering. Adapted from METR’s Time Horizon plot, including software verification benchmarks where AIs write code and then prove it correct. We plot only the time horizon for software implementation (not verification) for an an apples-to-apples comparison of how much functionality is implemented via each method of software development. lf-lean gives us an encouraging measurement of where verified software engineering capability is.
Кадр: Дмитрий Гордон / YouTube (автор канала внесен Минюстом РФ в реестр иноагентов; включен в реестр террористов и экстремистов Росфинмониторинга)
，推荐阅读雷电模拟器获取更多信息
Generative AI tools have a dial for bias, discrimination, and racism that you can’t control. From their source training material to the political alignment of major AI companies’ leadership to their adoption by fascist actors to consistently biased output, there’s a lot of red flags there. Their negative effects disproportionately affect women and historically marginalized groups, and contribute to a society-wide degradation in being able to understand what’s real.
更多精彩内容，关注钛媒体微信号（ID：taimeiti），或者下载钛媒体App
Opens in a new window