Uni agents set up by Microsoft were shocked by the 100 search results and took the first option. Malicious ad sellers can make all their complaints with fake reviews and scams. They're not ready for prime time, step-by-step human hands-independent AI shopping.
Microsoft included hundreds of AI agents who work as buyers and sellers, and then they saw people accepting the work they take on a daily basis. If the results are estimated at autonomous Ai market assistants, they should smoke.
With the Arizona State University on Wednesday, the promotion area of the market place research, 100 customer-side AQ agents as the order of the dinner and held about 300 business-side agents. Although the results show the promise of selfish representative business is not yet ripe.
When presented with 100 search results (too many for agents to handle effectively), leading AD models (how useful are the models)
All the agents have failed to make the necessary comparisons to make the first “good” option. This design, which the researchers call the “early question listener”, is captured by the researchers, which is 10-30x higher than the actual resolution.
But is there anything worse than that? Yes, malicious encouragement.
Microsoft combines psychological methods with psychological methods and offensive kick attacks, and six methods that provide social evidence. OPENI's GPT-4yo and its open-source model GPTOS-20b have all been successfully upgraded for follow-up. Alibaba's QWAN3-4b fell as the official's basic display techniques. Only Clued Solnet 4 resisted these display tests.
When Microsoft asks agents to work toward common goals, some don't know how to see effectively or how to coordinate effectively. Performance is obviously improved step by step with human guidance, but that defeats the whole purpose of independent agents.
So at least for now it seems you are better off doing your own shopping. “Representatives have to make human decisions, no one can replace them,” he said. Research suggests supervised automation, but agents receive human supervision and advice before making final decisions, and recommend that they hold and block recommendations.
The findings reach the deployment of autonomous market assistants such as Opena, Anthropic and others. The operation of Opena and brain clutch agents promise to browse websites and complete sessions without supervision. Microsoft's research is promising.
However, the fear of AI agents working in AIs is straining the relationship between AI companies and retail giants. Amazon recently sent a cease-and-desist letter to Perplexity AI, demanding it halt its Comet browser's use on Amazon's site, accusing the AI agent of violating terms by impersonating human shoppers and degrading the customer experience.
Confused back then, consumers should have the right to hire their own digital assistants instead of relying on controlled people.
An open simulation mode is now available to breed discoveries and reproduce the hells in their hellish activities.
