After a four-month interval, the QuestMindTM has received another major upgrade.

Recently, the China International Public Security Products Expo was held at the China International Exhibition Center in Beijing.

YITU TECH. showcased its latest QuestMindTM 4.5 at this year's Security Expo in Hall E1, with the theme "QuestMindTM, Expanding the Boundaries of Intelligence."

The newly upgraded QuestMindTM 4.5 has significantly improved its video content understanding capabilities. Not only can the model "see" and comprehend videos, but it now excels at quickly recognizing small targets and understanding relationships within scenes. This upgrade dramatically enhances its ability to understand target details and precisely deploy control strategies in complex environments.

Additionally, QuestMindTM 4.5 now supports the understanding of small target videos captured by drones, enabling rapid responses and precise interventions in dynamic and complex environments. This breakthrough has greatly facilitated the intelligent upgrading of urban traffic control and safety management, providing more efficient and accurate technological support for smart city management.

The QuestMindTM 4.5 also integrates natural language and visual search technologies, enabling deep integration of complex semantic understanding with fine-grained video comprehension.

By focusing on "user needs," the model deeply understands subtle contextual differences, allowing for rapid, detailed video content retrieval. The enhanced semantic understanding capability unlocks more open-ended semantic search scenarios, making it easier for users to interact via simple, natural language queries. This feature not only improves the efficiency of video content searches but also significantly speeds up the response times for various complex business operations.

In the process of large model deployment, one of the key challenges is converting business requirements into algorithmic tasks. With the release of QuestMindTM 4.5, Yitu introduces the Xiaoming AI Agent.

With a simple spoken or written task description, the Xiaoming AI Agent can automatically break it down into multi-condition algorithmic tasks, making the complex deployment of combined algorithms more accessible and less reliant on expert knowledge. This makes expert-level algorithm orchestration "democratized" for broader use.

The QuestMindTM 4.5 also includes an upgraded algorithm training and optimization process. The minimum number of positive examples required for algorithm iteration training has decreased by 75%. While maintaining the same training performance, the total sample size has been significantly reduced, and the algorithm iteration time has been shortened to minutes. The deployment time for new, real-world algorithms has been reduced to a matter of days.

This high-efficiency iteration capability is crucial for responding to sudden incidents and other time-sensitive situations, providing robust support and ensuring that urban management decisions are more intelligent and flexible.

 

At the 2024 Yitu Business Strategy Release Conference and Partner Summit on October 23, Yitu discussed the transition from the AI 1.0 era to AI 2.0.

In the AI 1.0 era, traditional deep learning relied heavily on large amounts of training data and scene coverage. The difficulty of acquiring data led to inefficiencies in AI algorithm production, making digital transformation in the security industry a challenge.

Now, with the onset of AI 2.0, cross-domain generalization and adaptability have significantly improved. Self-supervised learning models and multimodal AI models, driven by generalized pre-training and domain-specific fine-tuning, enable cross-domain intelligence capable of performing a variety of tasks.

The AI 2.0 technological revolution has brought four major core changes to the security industry:

  1. Self-learning and Post-Training: Pre-training + domain-specific fine-tuning improves adaptability across scenarios and domains. New long-tail algorithms can go from a monthly to a daily iteration cycle, boosting AI production efficiency.

  2. Contextual Understanding and Spatial Intelligence: The ability to perceive, locate, and assess in 3D space and 4D time-space, enabling full-scene, full-factor perception. AI is evolving from basic security to smart management and operations for production and safety.

  3. Unified Multimodal Data Representation: Fusion, cross-checking, and interaction between different types of data (e.g., text, visual, audio), enhancing sensory capabilities and creating a revolution in user interaction. Natural language interaction replaces "label filtering," improving the user experience.

  4. AI Agents: AI agents that analyze and understand causal relationships, use tools for reasoning and planning, and can shift from fast thinking to slow thinking. This shifts AI from being high-threshold and difficult to deploy to making algorithmic expertise accessible to everyone.

According to Yitu, the essence of AI 2.0 in intelligent security is that the marginal cost of producing long-tail algorithms is steadily approaching zero.

Yitu's Business Strategy for AI 2.0

In the AI 2.0 era, Yitu’s business strategy is clear:

  • Focus on Industry-Specific Large Models: Yitu will prioritize creating large models tailored to specific industry needs, addressing pain points in various sectors and offering fast scene unlocking with a broad scope.

  • Collaborative Solutions with Partners: Yitu is focused on providing a platform similar to PaaS (Platform as a Service) for large models, combining industry know-how with business closed-loop solutions and continuous operations.

  • Cost-Effective AI Products: Yitu emphasizes the importance of producing practical and cost-effective AI products. Ensuring affordability and scalability is the foundation of widespread AI adoption.

In addition to the model upgrades, Yitu also officially launched its MindVTM brand and a series of new products:

  • Zero-Shot Algorithm Generation: Turning user requests into algorithmic tasks with a single sentence.
  • Natural Language Orchestration of Combined Algorithms: Simplifying complex algorithms into atomic combinations.
  • On-Site Algorithm Tuning: Built-in AI data recommendation to fine-tune algorithms on-site for specific needs.

Yitu is committed to building an open and diverse cooperation platform, bringing together resources from various sectors to enhance market competitiveness, drive innovation, tackle challenges, and achieve mutual benefits.

You can copy the link to share to others: //dribblessportsbar.com/node/944