For the first time in Patronus AI Percival to help institutions monitor the agents of artificial intelligence on a large scale

Patronus ai A new monitoring platform today automatically determined failure in the systems of artificial intelligence agent, targeting the Foundation’s concerns about reliability as these applications grow more complicated.

The new AI Safety product in San Francisco, the new product, PersevalHe places himself as a first solution capable of identifying different failure patterns automatically in artificial intelligence agent systems and suggesting improvements to its treatment.

“Percival is the first solution to the industry, which automatically discovers a variety of failure patterns of agents, then systematically suggests reforms and improvements to address them,” said Anand Canapan, CEO and co -founder of Patronus AI, in an exclusive interview with Venturebeat.

Reliable crisis of artificial intelligence agent: Why companies lose control of independent systems

Institutions adopting artificial intelligence agents-capacity programs that can plan and carry out multiple multi-steps independently-independently-acceleration In recent months, creating new administrative challenges as companies are trying to ensure the work of these systems reliably.

Unlike traditional automatic learning models, these systems -based systems often include long sequences of processes where errors in the early stages can have severe consequences.

“A few weeks ago, we published a model that determines the possibility of agents’ failure, the type of effect that may occur on the brand, on the customer’s agent and such things,” said Canaaban. “There is a constant risk of mistake with the agents we see.”

This problem becomes particularly acute in multi -agent environments where various artificial intelligence systems interact with each other, making traditional test curricula increasingly insufficient.

Creating memory innovation: How to create the structure of AI Percival agent revolution in discovering mistakes

Perseval It distinguishes itself from other evaluation tools through its structure based on the agent and what the company calls “cross memory”-the ability to learn from previous errors and adapt to a specific workflow.

The program can discover more than 20 different failures across four categories: thinking errors, system implementation errors, planning and coordination errors, and field errors.

“Unlike LLM as a judge, Percival himself is an agent, and therefore he can track all the events that occurred throughout the path,” explained by Dishhan Disbandy, researcher at Patronus AI. “They can link them and find these errors through contexts.”

For institutions, the most urgent benefits appear to have been reduced to the correction time. According to Patronus, the first customers reduced the time the agent’s workflow analyzing from about an hour to one and 1.5 minutes.

Trail Benchmark reveals critical gaps in the capabilities of artificial intelligence

In addition to launching the product, Patronus launches a standard called Trail (tracking thinking and localization of the agents) To assess the quality of systems, problems can discover problems in the functioning of an artificial intelligence agent.

Search using This standard It revealed that even advanced artificial intelligence models are struggling with an effective tracking analysis, with the performance of the best performance is only 11 % on the standard.

The results emphasize the difficult nature of monitoring complex artificial intelligence systems and may help to clarify the reason for investing large institutions in specialized tools to oversee artificial intelligence.

AI leaders are adopted by Percival Foundation for Opinion Applications

It includes adoptions early The emergence of artificial intelligenceThat has almost sparked $ 100 million in financing It develops systems where artificial intelligence scientists can create and manage other agents.

“The recent penetration of the appearance-the factors that create factors-not only control the development of adaptive and self-generation systems, but also how these systems control and expand their scope with responsibility,” said Satia Nita, co-founder and CEO of AI, in a statement sent to Tanfurchry.

NOVA, another early client, uses a technology for a platform that helps large companies deport the old code through the SAP integration in which artificial intelligence works.

These clients depict the challenge aims to solve it. According to Kannappan, some companies are now managing agents systems with “more than 100 steps in one agent guide”, creating the complexity that exceeds what is possible for human operators efficiently monitors.

Artificial Intelligence Supervision Market is preparing for explosive growth with the spread of independent systems

The launch comes amid concerns about the increasing institutions regarding the reliability and rule of artificial intelligence. With companies spreading increasingly independent systems, the need for monitoring tools has grown proportional.

Canaaban pointed out that “the difficult thing is that the systems have become increasingly independent,” adding that “billions of code lines are generated daily using artificial intelligence,” which creates an environment where manual supervision becomes practically impossible.

The market of artificial intelligence monitoring and reliability is expected to expand significantly as institutions move from experimental publishing operations to important AI applications.

Percival integrates with several frameworks of artificial intelligence, including embrace Smolagentsand Pydantic aiand Openai Agent SDKAnd LinjshenWhich makes it compatible with different development environments.

while Patronus ai Not pricing pricing or revenue has not been revealed, the company’s focus on overseeing the level of institutions indicates that it defines itself for the safety market of AI Enterprise AI with high margin that analysts predict will grow dramatically with the adoption of artificial intelligence.

Daily visions about business use cases with VB daily

If you want to persuade your boss at work, you have covered VB Daily. We give you the internal journalistic precedence over what companies do with obstetric artificial intelligence, from organizational transformations to practical publishing operations, so that you can share visions of the maximum return on investment.

Read with us privacy policy

Thanks for subscribing. Check more VB news bulletins here.

An error occurred.