Users of ChatGPT can now secure restaurant reservations via AI agents, shop, and even compile lists of candidates for job openings. Starting Thursday, chatbots will function as personal assistants.
As stated by a US company, OpenAI has launched ChatGPT agents in regions beyond the EU. These agents merge AI research capabilities with functionalities that enable users to control various software like web browsers, document files, spreadsheets, and presentations.
This follows the introduction of similar “agents” by Google and other companies, which autonomously handle tasks such as creating travel itineraries and performing workplace research as interest grows in AI models adept at managing computer-based tasks by evaluating which software to use for switching between systems.
Niamh Burns, a senior media analyst at Enders Analytics, commented:
However, OpenAI recognizes that granting AI agents control over computer systems entails “greater risks in this model compared to the prior model.”
The goal is to assist users with daily tasks, but the potential risks prompted OpenAI to implement measures ensuring agents do not lead to biological threats.
“There is no definitive evidence that this model could significantly contribute to serious biological threats for beginners,” the company stated.
The system is designed to seek user approval before executing any harmful or irreversible actions. According to their blog: “You maintain control at all times. ChatGPT requests permission before undertaking any impactful actions.”
The rollout of this agent has raised questions about whether tech companies could monetize the service by guiding users to retail checkout. OpenAI CEO Sam Altman has suggested there may be a 2% fee on sales driven by the “Deep Research” software.
After the newsletter promotion
“These agents are independent of us,” Burns explained. Is there a commercial relationship where a brand is compensated for being highlighted by an assistant, or does it offer a unique product that sets it apart from the competition?
“As AI firms press for monetization of their products, we anticipate that certain advertising and sponsorship placements will become unavoidable.”
OpenAI clarified that the agent does not provide recommendations for sponsored products and has no intention of altering this policy.
In a recent software demo, users were prompted to check their Google Calendar and select an available weekday evening from 6 PM to 9 PM, then locate tables at Italian, sushi, or Korean restaurants with a minimum rating of 4.3 stars and offer them some options.
The task required 10 to 15 minutes, and like human assistants, users could intervene and redirect the AI agent’s focus. Likewise, agents can solicit clear instructions from users.
Another noteworthy risk involves agents potentially falling prey to malicious prompts hidden within the websites they explore, potentially passing a portion of user data to an agent.
OpenAI stated it has conducted numerous safety checks and trained its agents to reject specific suspicious requests, inclusive of bank transfer requests. The system will first be accessible to users subscribing to the “Pro,” “Plus,” and “Teams” versions of the model.
Source: www.theguardian.com
