This study is part of a growing warning organization about the dangers of deploying AI agents in real -world financial decision -making. Earlier this month, a group of researchers from multiple universities Debate LLM agents should not only be diagnosed with their high performance, based on their risk profiles. He says the existing benchmarks emphasize the accuracy and return -based matrix, which measures that an agent can perform well but ignore how far it can fail. His research has also shown that even high -performing models are more likely to be broken into anti -circumstances.
The team suggests that in the context of real-world financing, a small weakness-even a 1 % failure rate can also expose the system to systematic risks. They suggest that AI agents be “tested” tension before putting into practical use.
Assistant Professor Hanching Kao, who arrived at Emuri University, has noted that there are limits for pricing negotiation study. “These experiments were carried out in an artificial environment that may not fully achieve real -world negotiations or the complexity of consumer behavior,” says Cao.
Researchers and industry practitioners are experimenting with numerous strategies to reduce these risks, researchers say. These include improving the gestures given to AI agents, enabling agents to use external tools or codes to make better decisions, connecting multiple models to double checks, and the Fine Toning Model-all-on-based financial data on the domain-related financial data.
Many prominent AI shopping tools are currently limited to the recommendation of products. In April, for example, Amazon Launch what “buy for me” An AI agent that helps consumers find and buy products from other brands’ sites if Amazon does not sell them directly.
Although prices are rare in consumer e -commerce, it is more common in business transactions than business. Alibaba.com has developed a source assistant called Equivo built on its open source cue van models, which helps businesses find suppliers and research products. The company told MIT Technology Review It has no plans for price bargaining so far, citing high risk.
This can be a wise move. For now, PEI advises users not to treat AI shopping assistants as a helper tools-not standing for humans in decisions.
“I don’t think we are fully prepared to hand over our decisions to AI shopping agents.” “So use it just as an information tool, not a negotiator.”