ChatGpt Agent test shows the weird problems of this artificial intelligence!

OpenAI unveiled its latest smart agent, “ChatGpt Agent”; A tool that combines research and executive capabilities promises to do complex tasks instead of the user. But this new assistant has strange problems: it requires human permission to do any important job, and in the early experiments, it has a great deal of mistakes and performance, showing that there is still a long way to go to real independence.

Openai has introduced a new and integrated tool directly in the ChatGPT environment by combining its two previous smart factors, Operator (for the browser) and Deep Research (for multi -stage research). Using a “virtual computer”, this smart agent can do tasks such as calendar review, travel planning, online shopping or analytical reports.

Strange ChatGpt Agent Issues

Despite these attractive capabilities, a fundamental restriction has questioned the usefulness of this tool. According to Openai’s official announcement, the smart agent “will be allowed to allow the user before any important action.” This means that you can’t handle it and get up from the system. For every important step, from booking tickets to online shopping, your final presence and confirmation are required.

According to Weird’s analysis, this dual approach reveals an important fact:

• From a safety point of view: This is a clever action. Since artificial intelligence is prone to mistake or fall into the trap of cyber attacks (such as a periphery injection), this human oversight prevents financial or security catastrophes.
• From a practical point of view: This continuous need for human intervention questions the main purpose of a “automation” tool. This intelligent agent is caught in a strange limbo: too Powerful That cannot be abandoned and so much False It is not possible to handle things alone.

The performance of this smart assistant in the early experiments has not been so promising; Experts say by testing this smart factor to order a few cup cakes by this intelligent agent An hour It took a long time.

Also in the video introduction of this product, when ChatGpt Agent is asked to plan a trip to visit all the US Premier League stadiums, it produces a map that holds one of the stops in the middle The Gulf of Mexico Shows. These gross errors, even ignored by the provider’s own, show that the technology is still far from reaching an acceptable level of confidence and efficiency.

This new feature first for Pro subscription users with limitation ۱۰۰ Permapt per month It comes. Subscribers Plus And Team Also soon, but with much more restrictions (only one tenth of Peru users) will have access to it. No time has been announced to free users to release this feature.