While AI business and automation innovation carriers remain to proclaim the turbulent modifications AI– and particularly AI representatives– are offering services, the results of a research project at Carnegie Mellon can give convenience to people that just see work loss and catastrophe in the future. While automation is certainly making companies a lot more effective, stress over makers taking control of totally are misguided– at the very least now in time.
First reported by Organization Expert, the CMU group produced a business and staffed it totally with AI Professionals.
” In this paper, we present TheAgentCompany, an extensible standard for examining AI representatives that connect with the globe in comparable methods to those of an electronic employee: by surfing the Internet, composing code, running programs, and connecting with various other colleagues,” the scientists created. “We developed a self-supporting atmosphere with interior website and information that simulates a little software program firm atmosphere, and develop a selection of jobs that might be done by employees in such a business.”
While the representatives– standing for LLMs from a few of one of the most widely known AI business consisting of Google, OpenAI, Anthropic and Meta– promptly finished a few of the jobs they were appointed, the best of them were just able to finish 24 percent totally autonomously. Representatives additionally regularly misunderstood discussions with associates or would not act on essential instructions, too soon noting the job full.
Regardless Of that, the record claimed brand-new LLMs are making substantial progression and strategy to utilize the existing research as a standard, going back to the try out advanced versions.
” Not just are they ending up being increasingly more qualified in regards to raw efficiency, however additionally a lot more inexpensive,” the record ended. “Open-weights versions are shutting the void in between exclusive frontier versions also, and the more recent versions are obtaining smaller sized however with equal efficiency to previous big versions, additionally showcasing that effectiveness will certainly even more boost.”
发布者：Dr.Durant，转转请注明出处：https://robotalks.cn/ai-agents-fail-in-experiment-but-show-much-promise-say-researchers/
