OpenAI has introduced ChatGPT Agent, designed to automate various tasks by utilizing a virtual computer. This tool combines the functionalities of the Operator and Deep Research agents, enabling users to manage tasks via ChatGPT's interface. While it can assist in organizing schedules and conducting research, it necessitates user intervention for significant tasks to ensure safety. This requirement for approval reflects ongoing concerns over AI reliability and trustworthiness, highlighting the technology's current limitations in balancing autonomy and control.
"ChatGPT requests permission before taking actions of consequence, meaning that for any actually important task, you can't just walk away after setting things into motion."
"From a safety point of view, this is unequivocally a good thing, given that AIs are extremely prone to making mistakes."
"This intervention underscores just how untrustworthy this technology remains, which by extension incapacitates its usefulness, suspended in this awkward limbo in which it's both too dumb and too powerful to just let loose."
"The new agent synthesizes the capabilities of its Operator agent and its Deep Research agent, allowing you to access the tool from the comfort of the chatbot's UI."
Collection
[
|
...
]