Former Google researchers have made a new type of artificial intelligence agent

New type to artificial intelligence Agent, the trainer to understand how to build programs by overcoming the company’s data and learn how to lead to a final product, can be a more capable programs and a small step towards artificial intelligence more intelligent.

The new agent, called ASIMOV, has been developed by Reflection, a small but ambitious startup company confused by Google International. Asimov reads the code as well as emails, SLACK messages, project updates and other documents with the aim of learning how to lead all this together to produce a piece of final software.

Reflection’s ultimate goal is to build excellent artificial intelligence – something other than artificial intelligence is that they work for it. Meta recently created a new laboratory for my calligrapher, Promising Researchers interested in joining his new efforts.

I have visited the Reflection headquarters in the Williamsburg neighborhood in Brooklyn, New York, by directly from the Pickable Bicky-Glid, to find out how to plan meditation to reach experts before competition.

The company’s CEO, Misha Laskin, says the ideal way to build Ai Supersmart agents is really progressing, because this is the most natural way for them to interact with the world. While other companies build agents who use human user facades and Browse the webLaskin, who previously worked on Gemini and agents at Google DeepMind, says this barely comes naturally to a large language model. Laskin adds that teaching artificial intelligence to understand software development will also lead to the production of more useful coding assistants.

Laskin says ASIMOV is designed to spend more time reading the code instead of writing it. He told me: “Everyone really focuses on generating the code.” “But how to make the agents useful in preparing the team is not really resolved. We are at this semi -self when the agents have just started work.”

ASIMOV actually consists of several smaller factors inside the trench coat. All agents work together to understand the code and answer user inquiries on this topic. The smaller agents recover the information, and the biggest thinking agent collects this information in a coherent answer to inquire.

The reflection claims that ASIMOV is already considered to be outperforming some of the AI leadership tools through some measures. In a poll conducted, the company found that developers working on large open source projects and who have posed favorite questions from ASIMOV 82 percent of time compared to 63 percent of the HotHROPIC code that runs the Sonnet 4 model.

Daniel Jackson, the computer world at the Massachusetts Institute of Technology, says the Reflection approach appears to be promising due to the broader scope of information collection. However, Jackson adds that the benefits of the approach to be seen, and that wiping the company is not sufficient to persuade it with wide benefits. It indicates that the approach can increase the costs of the account and potentially create new safety problems. “He will read all these special messages,” he says.

Reflection says that the multi -factor approach reduces account costs and that it uses a safe environment that provides more safety than some traditional Saas tools.