Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
Join a reliable event by enterprise leaders in about two decades. VB Transform, Real Enterprise AI strategy brings together people who build. Learn more
Many enterprises never do the AI Agent Development Efforts Never produce and not because technology is not ready. Problem, according to DatabricksCompanies still rely on hand evaluations with a slow, inconsistent and scale process.
Today, the information was launched at the + AI summit, the Databrices Mosaic Agent brick as a solution to this problem. Is built on technology and stretches Mosaic ai agent frame The company announced in 2024. Simply put the AI agents to have a real world effect, it is no longer good enough to be able to build.
The Mosaic Agent Bricks Platform automates an agent optimization using a number of research-supported innovations. There is integration between major innovations Tao (Timely adaptation optimization), which provides a novel approach to AI tuning without the need for labeled data). Mosaic agent bricks also create special synthetic information with the area, create aware of the assignment and optimize the balance from the quality of quality.
Basically the aim of the new platform is to solve an issue available with the EI Agent development efforts of the existing AI agent.
“They did not have a way to evaluate the blind, the head of the main technology of the main technology of the General Technology of the Hanlin Tang, Databricks Nine Networks,” Most of them, manually, hand vibe, but it does not give confidence in relying on production. “
Enterprise AI Production Scale from Research Innovation
Tang was a mutual co-founder and CTO earlier obtained by data In 2023 to $ 1.3 billion.
In the mosaica, most of the research updates were not necessarily an effect on the enterprise. Changed after all obtained.
“The great light lamp for me was the time we started our product in our product, and immediately, when we used it as a client of thousands of enterprises,” Tang.
In contrast, the mosaic will spend months trying to take a handful of enterprise to test the products before obtaining. Mosaic’s integration into the database of Mosaic has provided direct access to the research teams of the research team and revealed new areas to investigate.
Contact this enterprise has discovered new research opportunities.
“This is only in contact with enterprise customers, you are working deep with them that if you solve the really interesting research problems.” Agent brick …
Agent AI Evaluation Crisis Resolution
Enterprise groups face an expensive test and error optimization process. Your task is to get an expensive guess game, without custom test information, without custom test information. Follow quality sliding, cost and missed periods.
Agent Brick automates the entire optimization pipeline. The platform receives a high-level task description and enterprise information. Automatically manages the rest.
First, it creates special evaluation and LLM judges in the tasks. Next creates synthetic information that reflects the customer’s data. Finally, the best configuration is searching for optimization methods to find.
“The customer describes the problem at a high level and does not enter low-level details because we take care of them,” Tang said. “The system creates synthetic information and builds special LLM judges inherent in each task.”
The platform offers four agent configurations:
- Information extraction: Converts documents (PDFs, emails) into structured data. A use box can be retail organizations using the supplier to provide product details from supplier PDFs, even complex formatting.
- Knowledge assistant: Provides accurate, referenced answers from enterprise information. For example, production technicians can get immediate response from the maintenance of production through binders.
- Special llm: Text transformation tasks (generalization, classification) will handle. For example, medical organizations can make models of patient records for portic work streams.
- Many agent controller: Many agents are orchestra for complex work flows. A use is financial services firms that can coordinate agents for instance, intent detection, document acquisition and compliance checks.
Agents are excellent, but do not forget the data
The creation and assessment of agents is the main part of the AI company’s ready, but not the only part.
Databricks, mosaic agent bricks, like a consumer layer of AI sitting with a unified data stack. Data + At the AI summit, announced the general availability of data in data GAKEFLOW DATA ENGINETING The previewed platform in the first 2024.
GEAWFLOW solves the problem of data preparation. Three critical information that previously requires separate tools combines engineering journeys. Gaining is obtained from both structural and unstructured data in data. The transformation ensures effective data cleaning, retraining and preparation. The orchestra manages production work and planning.
Workflow connection is directly: Lake develops enterprise information through a single unit admission and transformation, then the agent brick builds an optimized AI agents in this prepared data.
“We can help enter the information on the platform and then you can do ML, BI and AI analysts,” Bilal Aslam, “Bilal Aslam, CEO Bilal Aslam at Databreat in Venturebeat.
The mosaic agent brick, which is released from the data adoption, also benefits from the management features of Databricks’s unity catalog. This includes access control and information lineage tracking. This integration ensures the respective of the management of enterprise data without additional configuration of agent behavior.
The agent from human opinion eliminates the filling request
One of the general approaches to lead AI agents today is to use a system request. Tang, all kinds of management of all kinds of management of users hopefully the agent will follow him with a desire to follow a desire.
The agent presents a new concept called bricks – learning agent from human feedback. This feature automatically regulates system components based on natural language management. The tangin solves that the quick filling problem. According to Tang, the emergency charging approach often fails, because there are more than one component in need of agent systems regulation.
The learning agency from the human opinion is automatically commenting on the natural language management and regulating the relevant system components. Approach strengthening mirrors from human review (Rlhf) However, the agent operates in the system level than the individual model weights.
The system manages two main problems. First, the natural language leadership can be indefinitely. For example, ‘Respect your Brand’s voice’ What does it actually mean? Second, there are numerous configuration points in agent systems. Teams fight to determine which components need to be adjusted.
The system eliminates the estimates where the agent components need to be adjusted for special behavioral changes.
“We believe that agents will help further manage.”
Technical advantages in existing frames
Today, there is no shortage of agent on the market agent AI development frames and tools. There are tools between the list of growing vendors Langchain, Microsoft and Google.
Tang, Mosaic agent claimed what the brick was and what optimize. Instead of using manually configuration and tuning, agent brick automatically combines multiple research techniques: TAO, context learning, emergency optimization and delicate adjustment.
When it comes to the agent agent for the agent communication, today there are several options on Google, including the market Agent2Agent Protocol. According to Tang, Databicks are currently investigating various agent protocols and does not remain committed to the unified standard.
Currently agent brick, agent-agent communication through two initial methods:
- To expose agents such as endpoints that can be twisted on different protocols.
- Using multiple agent supervisors with MCP (model context protocol).
Strategic results for enterprise decisions
It is important to have the right technologies to assess quality and effectiveness for businesses wishing to lead the road in AI.
The placement of the Qur’an agents will not lead to an optimal result and there will be no agents without a solid data fund. When reviewing agent development technologies, it is important to have the right mechanisms to evaluate the best options.
The agent who studied the agent without approaching human opinion is also noteworthy for the entity who helped the best result of the EI.
The evaluation infrastructure of this development is no longer a blocking factor for enterprises looking for leadership in the AI agent. Organizations can focus on the authentication of the rules of use and data preparation of optimization frames.
Source link
