A Hugging Face team has released a “agent” that uses the computer “agent” computer. But be expected: it is quite slow and occasionally makes mistakes.
Hugging the face of Face, called Open Computer Agent, is accessible via the web and can use a preloaded Linux virtual machine with different applications, including Firefox. Similar to the Openai operator, it is possible to induce Open Computer Agent to complete an activity – let’s say: “Use Google Maps to find the headquarters of Hugs in Paris” and sit down while the agent opens the necessary programs and understands the requested steps.
Open Agent Computer can manage simple requests quite well. But the most complicated ones, such as the search for flights, triggered him in Techcrunch tests. Open Computer Agent also runs into Captcha test that is unable to solve.
You will also have to wait in a virtual tail to use an open computer agent – a tail of a few seconds or minutes, depending on the question.
Obviously, the goal of the embraced team was not to build a cutting -edge agent at the forefront. Rather, they wanted to demonstrate that open artificial intelligence models are becoming more capable and cheaper to work on the cloud infrastructure.
“As the vision models become more capable, they become able to feed workflows complex agents”, Aymeric Roucher, a member of the Hugging Face team, wrote in an X post on X. “(some of these models) support integrated land, that is (the) ability to identify any element in an image with its coordinates, (E) (E) a virtual machine). “
While it is far from perfect, agent technology is attracting growing investments while companies try to adopt it to increase productivity. According to a recent KPMG survey, 65% of companies are experimenting with artificial intelligence agents. Markets and markets plan that the segment of agents will grow from $ 7.84 billion in 2025 to $ 52.62 billion by 2030.
Techcrunch event
Berkeley, ca.
|
June 5th
Book now
