OpenAI has revealed it's new 'Agent' tool which boasts 'a new level of capability' for AI tools, but you might not want to dive in head first as ChatGPT's creator has revealed the clear risks and dangers present in the new software.
Sam Altman is never afraid to share the weaknesses of his own technology, as only recent was he left shocked that people trusted ChatGPT amid clear hallucinations and worryingly sycophantic behavior.
This has continued with his company's latest piece of tech too, as while he's clear about the impressive features that 'Agent' will bring to its users life, he's equally as open about the risks they'll face.
What is OpenAI's new Agent tool?
As shared by Altman in a new lengthy post on X, 'Agent' is an enhanced form of ChatGPT that allows users to achieve more complex and long form tasks compared to the standard chatbot form.
Advert
It uses it's own computer to complete tasks, letting it 'think' for far longer and complete complex chains where it combines several different actions into one lengthy process.
"For example, we showed a demo in our launch of preparing for a friend's wedding: buying an outfit, booking travel, choosing a gift, etc," Altman explains. "We also showed an example of analyzing data and creating a presentation for work."
Instead of having to input a new prompt for every new task you want an AI assistant to complete, Agent can now work on larger multi-step projects, almost akin to a personal secretary.
While that all sounds very impressive, there remain clear dangers in its current form, and Altman has issued strong warnings to anyone looking to pick up Agent right away as there are some things you'll want to steer clear of.
What are the risks when using Agent?
OpenAI is taking an 'iterative development' approach to Agent, and while it has implemented and built significant safeguards for the new technology, it adds that the team "can't anticipate everything."
Altman illustrates that he "would explain this to my own family as cutting edge and experimental; a chance to try the future, but not something I'd yet use for high-stakes uses or with a lot of personal information until we have a chance to study and improve with it in the wild."
It appears that Agent is still susceptible to being 'tricked' by bad actors, leaving it vulnerable to giving your private information away. As a consequence, Altman has urged used to give the tool the minimum required access to your data.
"For example, I can give Agent access to my calendar to find a time that works for a group dinner," Altman writes, "but I don't need to give it any access if I'm just asking it to buy me some clothes."
Another example he provides is asking Agent to go through your emails and 'address them' without any follow up questions. This could lead to the AI clicking on a cybersecurity scam email that a human would be able to spot but an AI would leave itself vulnerable to, and it would likely hand over your personal data without hesitation.
"We think it's important to begin learning from contact with reality, and that people adopt these tools carefully and slowly as we better quantify and mitigate the potential risks involved," Altman urged. "As with other new levels of capability, society, the technology, and the risk mitigation strategy will need to co-evolve."