Extracting Insights from Data: The Latest Technologies in Automated Dashboard Generation

In today's world, where corporate digital transformation is accelerating, rapid, data-driven decision-making is the key to establishing a competitive advantage. However, finding truly valuable insights from the vast amounts of data accumulated daily and visualizing them in an easy-to-understand format has been a challenge requiring specialized knowledge and significant time.

This article introduces an end-to-end framework that utilizes multi-agent LLMs to extract domain knowledge from data and bridge it to advanced analytical perspectives and visualization.

This technology enables a process that extracts business-critical insights from raw data and automatically generates effective dashboards. We will examine, with concrete examples, how the fusion of automated analysis and visualization can transform enterprise data utilization.

Listen to easy-to-understand audio content on Spotify: "Rami's AI Tech Radio"

A World Economic Forum survey ranked "curiosity and lifelong learning" as the 5th most important skill needed in 2030.Curiosity is not a hobby, but an economic necessity.
I explain this in detail in my book, "The Ultimate Skill for the AI Era: Curiosity Power."

Preface

It has long been said that "data is the new oil," but the process of refining that oil and utilizing it as energy remains complex. Especially in business settings, there is a demand to distill the results of data analysis into a form that everyone can understand—a "dashboard"—and connect it to rapid action.

However, this path from "data to dashboard" has never been easy, even for analysts with specialized skills.

Although visualization and BI tools have become widespread, they often still face challenges when it comes to deepening insights beyond the "accuracy of the outputted charts and tables." In this article, we introduce a cutting-edge framework using the concept of multi-agent LLMs, covering everything from the principles of the technology to examples of its application in business.

Current Status and Challenges of Data Analysis and Visualization

In modern society, data is being generated and accumulated in various fields, including corporate activities. Appropriately analyzing this data and using it for business decision-making has become an essential element for companies to maintain competitiveness and continue growing.

In particular, in an era of intense market fluctuations and diversifying customer needs, it is no exaggeration to say that rapid and accurate data-driven judgment determines the future of a company.

1. The Importance of Data Analysis in Business

The data collected by companies is diverse, including customer purchase history, website access logs, operational status obtained from sensors, and employee performance data. When properly processed and analyzed, this data creates value in the following ways:

As shown, data analysis plays an important role in every aspect of corporate activity. And for stakeholders to intuitively understand and share the insights gained from analysis, "data visualization" is an extremely effective means.

Dashboards visually display important KPIs (Key Performance Indicators) and data trends using graphs and charts, conveying complex information in an easy-to-understand way and supporting rapid decision-making.

2. Challenges of Conventional Data Analysis and Visualization Processes

While the importance of data analysis and visualization is increasing, there are several challenges in conventional processes.

As companies steer toward becoming data-driven organizations, transforming the vast amount of information collected daily into useful insights has become an urgent priority. The data analysis process can be broadly divided into the following phases.

However, many organizations face the challenge that not only is time and effort spent on "preprocessing," but "visualization and analysis" also require expert knowledge, and because the person in charge's subjectivity tends to intervene, the quality of the final output varies.

Furthermore, even now that the use of BI tools has become common, how to incorporate business-specific domain knowledge into data analysis and reflect it in visualization remains a major theme. Questions such as "What indicators should we focus on?" and "From what angle should we look at graphs to gain discoveries that lead to strategy?" have often required subjective judgment and rules of thumb.

Against the backdrop of these challenges, multi-agent LLMs have been attracting attention in recent years. Conventional LLMs have been applied to a wide range of uses, from casual Q&A to text generation, due to their massive parameter counts and natural language understanding capabilities. However, in business data analysis and visualization, methods involving "specialized role division" and "incorporation of domain information" are seen as even more effective.

In this article, while building on the key points of the paper "Data-to-Dashboard: Multi-Agent LLM Framework for Insightful Visualization in Enterprise Analytics," we will detail how multi-agent LLMs make the following possible.

As a next-generation approach to accelerate corporate decision-making, we will look at the benefits and usage methods of multi-agent LLMs.

What is a Multi-Agent LLM?

1. Differences between Conventional LLMs and Multi-Agent LLMs

LLMs are AI models that can handle natural language very fluently. While they can handle a wide range of tasks such as document summarization, Q&A, and reasoning, as the application domain becomes more complex, the following challenges emerge.

Therefore, a multi-agent approach has been proposed. This is a configuration where multiple "expert agents" are placed within the LLM, each responsible for a specific role or purpose.

For example, roles are clearly divided, such as one agent performing "domain determination" and another agent "selecting the graph type suitable for visualization." Each agent has partial prompts and memory, and they work together to produce high-quality final output.

2. The Importance of Domain Knowledge

One of the reasons why multi-agent LLMs are particularly attracting attention is the utilization of domain knowledge. Domain knowledge refers to information regarding a company's industry characteristics, the products or services it handles, or the priorities of financial and sales indicators. Whether or not the model understands these significantly changes the direction of analysis and the quality of the insights obtained.

For example, in the retail industry, indicators such as "average customer spend," "repeat rate," and "inventory turnover rate" increase in importance, while in the SaaS business, "retention rate," "upselling," and "MRR (Monthly Recurring Revenue)" become important. To make conventional LLMs understand these, a large amount of additional training or careful prompt design was required.

On the other hand, with multi-agent LLMs, it is possible to build a mechanism that sets up an agent dedicated to "domain detection," identifies keywords and characteristics of the relevant industry or business type, and passes that information to the next analysis agent. This reduces the risk of pursuing incorrect analysis policies and makes it possible to smoothly pick up key points that align with the business.

3. End-to-End Pipeline

In the framework covered in this article, a series of flows such as "Raw data -> Domain detection/analysis -> Visualization -> Insight extraction" are executed in stages on a single system.

　By placing dedicated agents at each stage and having each pass their results to the next, we replicate a workflow similar to the process used by human business analysts to derive insights.

　For example, if the domain detection agent determines that "this data is related to retail, and the key metrics to focus on are inventory turnover and average customer spend," the subsequent analysis agent will generate insights focused on "inventory-related figures and sales trends."

　Following this, the visualization agent makes specific suggestions, such as "output monthly inventory trends as a line graph and overlay a bar chart for sales revenue," and then proceeds to generate explanatory text and captions.

Overview of the New "Data-to-Dashboard" Framework Proposal

Here, we will organize and explain the key points of the paper, "Data-to-Dashboard: Multi-Agent LLM Framework for Insightful Visualization in Enterprise Analytics."

　Based on the challenges in data analysis and visualization, this paper proposes a new framework called "Data-to-Dashboard." This framework aims to achieve an end-to-end process, from raw data ingestion to the automated generation of insightful dashboards, through the collaboration of multiple AI agents centered around an LLM.

　This framework is specifically designed to support deeper insights and faster decision-making in enterprise business analytics environments.

Framework Overview and Objectives

　The primary goal of the "Data-to-Dashboard" framework is to "generate domain-knowledge-based insights from raw data and automatically create meaningful data visualizations (dashboards) based on those insights."

　In conventional data analysis systems, particularly those based on question-answering (QA), users were required to have clear questions or hypotheses in advance. However, this new framework does not rely on such prior inquiries; instead, it aims for the LLM to autonomously explore datasets and discover insights that the user might not have anticipated.

　The paper suggests a contrast between existing approaches and this proposal. In existing approaches, LLMs generate context-aware outputs based on raw data and metadata (contextual information), but they often risk overlooking deeper value embedded within the raw data.

　In contrast, this proposal aims to extract more essential and profound insights that are not dependent on context by actively utilizing domain knowledge.

　The "Data-to-Dashboard" framework adopts an architecture known as a "multi-agent system." This is a system where multiple autonomous "agents" work in coordination to achieve specific tasks.

　Each agent possesses specific roles and expertise, and they perform more complex and advanced tasks as a whole by exchanging information and passing on processing results to one another.

　In this framework, the LLM functions as the brain for each agent. Specifically, it consists of the following groups of specialized agents:

　By having these agents collaborate, the framework attempts to simulate a flexible and deep analysis process, as if a team of human business analysts were performing the work.

Two Main Stages of the Proposed Method

　The processing of the "Data-to-Dashboard" framework is broadly composed of the following two stages.

Stage 1: From Data to Insights

The input for this stage is raw data only, and no additional contextual information is required. As output, the dataset's domain name, domain-specific concepts, and analytical insights are obtained. This process is executed by the following modularized agents.

Data Profiler
Just as a human analyst would first try to grasp an overview of an unknown dataset, this agent automatically constructs a statistical summary of the dataset.

Specifically, it infers the data type of each column, value ranges and units, functional dependencies between columns, and potential keys. This deepens the structural understanding of the data before analysis begins, thereby increasing analytical efficiency. The paper states that by using a prompting technique called "Tree-of-Thought (ToT)" for this process, more detailed profiling is achieved.

Domain Detector
Based on the structural profile information created by the Data Profiler, it determines which business domain the dataset belongs to. In doing so, rather than relying on a fixed classification system, it refers to external knowledge sources like Wikipedia to flexibly generate domain labels (e.g., "Financial Services," "Customer Relationship Management," etc.) and their concise definitions.

This allows the system to handle a wide range of datasets without being tied to specific industries or predefined ontologies. In the future, the integration of a multi-round verification mechanism using LLMs is also being considered to enhance the consistency and accuracy of the generated domain labels.

Concept Extractor
Based on the identified domain labels and the data's metadata (structural information), it identifies prominent concepts that are likely to be important in downstream analysis. These concepts are expressed as natural language phrases such as "Monthly Active Users," "Unit Cost," and "Processing Latency," and are extracted in a way that corresponds to the domain theme and the results of the data profile.

This agent ensures that the extracted concepts are not only relevant to the dataset's business theme but are also practical for the analysis tasks performed by the subsequent analysis generator.

Through this Stage 1, raw data is structured, endowed with business context, and sublimated into "insights" analyzed from multiple perspectives.

Stage 2: From Insights to Charts

The role of this Stage 2 is to convert the analytical insights generated in Stage 1 into effective visualizations (charts and graphs) suitable for the domain. In this process, the "Tree-of-Thought (ToT) reasoning framework" is employed.

ToT was chosen to enable structured, multi-step reasoning in complex decision-making processes such as selecting chart elements, and to simulate the process of expert deliberation. Unlike approaches that determine visualization choices in a single pass, ToT facilitates the explicit consideration of alternatives and their impact on the domain.

Specifically, the system evaluates competing visualization strategies against domain requirements through a "consensus-building mechanism by three experts," discusses the effectiveness of different chart types, and scrutinizes visual encoding choices before deciding on the final representation.

This deliberative process leads to a consensus that identifies the optimal visualization type, the reasons for its selection, a narrative of key insights, and recommended annotations that emphasize domain importance.

As a result, the system maintains the context of the domain and insights throughout the visualization pipeline, ensuring that the generated charts function not just as data representations, but as a means of conveying substantive domain knowledge.

Interestingly, the paper states that "it was found that the higher the quality of the generated insights, the more it conversely affected the accuracy of chart generation." In particular, it was noted that when domain insights from Stage 1 were effectively communicated to Stage 2, it tended to become more difficult to generate accurate charts, especially legends. This suggests the difficulty of distilling highly complex insights into simple and accurate visuals, which can be considered one of the points for future improvement.

In this way, the "Data-to-Dashboard" framework aims to extract valuable insights from raw data and express them as effective dashboards, with multiple expert agents collaborating and running self-improvement loops. In the next chapter, we will look at how well this framework actually performs through concrete experimental results.

The key here is not the superficial axis labeling performed by standard "automated graphing" tools, but rather generating annotations and explanatory text based on business meaning and insights. A major strength of multi-agent LLMs is the ability to add a narrative, such as "simultaneously visualizing the correlation with inventory costs to provide suggestions for future ordering strategies," rather than simply labeling a chart as a "monthly sales bar chart."

Comparison with Conventional Approaches

Differences from QA-based Systems

Many conventional chart generation and business analysis support tools have functioned by responding to user questions (e.g., "What are the top 5 products by sales in March?"). While such "QA-pair" type tools offer a certain level of convenience, they have the following limitations:

The framework proposed in the paper differs significantly in its ability to automate open-ended exploratory analysis. Even without a specific question, it autonomously extracts anomalies and business metrics inherent in the data and visualizes or describes them, allowing it to pick up insights in areas where the user has not set up questions in advance.

Comparison with Manually Created Visualizations

As seen on platforms like Kaggle, reports where skilled data analysts manually create graphs and add commentary are extremely useful. However, manual work often leads to the following issues:

The framework provided by multi-agent LLMs can incorporate measures to enhance decision-making transparency (such as publishing prompts for each agent or using self-reflection modules), which has the advantage of allowing the visualization process itself to be verified and re-executed later. This can also be an effective mechanism when large companies attempt to standardize data utilization.

Experiments and Evaluation Methods

Evaluation of Text Output: G-Eval

In the paper, G-Eval is adopted as a method for evaluating insights automatically generated by language models. G-Eval was originally a framework for evaluating the quality of natural language generation, but in this study, it has been customized with a focus on the following perspectives:

These are perspectives also valued in actual business analysis, and a feature of this approach is that it can grasp how much analysis that extracts practical value has been performed, rather than just accuracy (whether the numbers are correct).

Evaluation of Visualization

The quality of a visualization cannot be judged solely by whether the graph type is appropriate. The insights a human gains from looking at the graph, and the consistency of the text annotations that support them, are also important. The paper verifies the following points using comparisons with visualizations created by Kaggle users and subjective evaluations by human experts (user studies):

Because visualization by multi-agent LLMs inevitably depends on the code generated from prompts (such as Python's matplotlib or Plotly), generation errors may occur. On the other hand, the paper reports that it has the advantage of being able to automatically propose complex graphs that might be overlooked manually.

Concrete Examples

Here, we will briefly summarize the cases introduced in the paper and present examples assuming actual business scenarios.

Example of Call Center Operations

When analyzing data from a company's call center, one might traditionally look at disparate metrics such as "response rate," "average wait time," and "customer satisfaction" individually. By introducing a multi-agent LLM framework,

Such output automatically derives complex visualizations that tend to be cumbersome when done manually, and because it adds annotations based on domain knowledge, management can also quickly grasp the situation.

Example of Marketing Data

As mentioned at the beginning of the paper, marketing activities involve a wide variety of perspectives, such as "new customer acquisition channels," "repeat purchases," and "CVR (Conversion Rate) by promotion." The usefulness of a multi-agent LLM lies in its ability toautomatically link multiple channels and customer attributes contained in the data to derive integrated insights.in doing so.

For example, one could consider a form that simultaneously suggests conflicting metrics—such as customers acquired via SNS having high initial purchase amounts but low retention rates, while customers acquired via email campaigns have small initial purchase amounts but high LTV (Customer Lifetime Value)—and highlights them through multiple visualizations. This could significantly speed up the parts where marketing staff previously had to use trial and error for segmentation and analytical perspectives manually.

Discussion

In the previous chapters, we have looked in detail at the mechanisms of the "Data-to-Dashboard" framework and the experimental results demonstrating its effectiveness. In this chapter, we will discuss what these research findings mean and what kind of future they might bring to the fields of data analysis and business intelligence.

What the research results tell us: The key is the automatic utilization of domain knowledge

One of the most important findings of this research is that"explicit domain identification and the utilization of domain knowledge significantly improve the quality of generated insights." It was shown that by not just having the LLM analyze data, but by making it think about "what field does this data belong to?" and incorporating that field's specific knowledge and perspectives (e.g., customer lifetime value and channel analysis for marketing, or risk and profitability for finance) into the analysis, deeper and more business-relevant insights can be obtained.

This reaffirms the importance of "domain knowledge" that human analysts cultivate through years of experience and learning, while suggesting the possibility that AI can replace or supplement a portion of it.

Traditionally, even if AI-based data analysis was good at statistical pattern extraction, "contextual understanding"—what those patterns mean in a business sense—has been a challenge. Approaches like "Data-to-Dashboard" can be seen as attempts to bridge this gap in contextual understanding by leveraging the LLM's broad knowledge base and reasoning capabilities.

Furthermore, the multi-agent system approach, where multiple specialized agents cooperate and deepen their analysis through self-reflection, has been shown to be more effective in producing high-quality results than having a single LLM process complex tasks all at once.

This is similar to the human organizational problem-solving process, where complex problems are broken down into small tasks and each expert (agent) takes responsibility for solving them.

The value brought by the automatic utilization of domain knowledge

AI systems that can automatically utilize domain knowledge have the potential to bring the following values to companies.

How will the role of business analysts change?

As such AI technologies become widespread, it is expected that the role of business analysts will also evolve. While AI will take over routine analysis and dashboard creation, human analysts will be required to take on more advanced and strategic roles, such as the following:

Defining business problems and instructing AI
The role of defining which business problems to solve, what data to use, and what analysis the AI should perform, and then providing precise instructions. The ability to 'ask the right questions' to maximize AI's capabilities will become even more important.

Evaluating and interpreting AI-generated insights
The role of critically evaluating whether the insights and dashboards generated by AI truly align with the business context and are valid, and determining how to translate those results into concrete business actions. It is necessary to avoid taking AI 'answers' at face value and instead combine them with human expertise to make final decisions.

Complex decision-making and strategy formulation
Complex decision-making that requires ethical judgment, creativity, and a long-term vision—things AI cannot fully handle—as well as strategy formulation based on these, will continue to be an important role for humans. A collaborative relationship where humans use AI as a powerful 'analysis partner' while retaining final control will be essential.

AI model management and tuning
The role of appropriately managing AI models according to the company's business and data characteristics, and performing tuning as needed. Specialized knowledge to understand AI behavior and maintain or improve its performance may also be required.

Communication and storytelling
Communication skills to convey AI analysis results in an easy-to-understand manner to those without specialized knowledge, such as management and members of other departments, and the ability to weave stories from data will become increasingly important.

AI will not take away the jobs of analysts, but rather shift them toward more advanced tasks and act as a force that expands their capabilities.

Updating and managing domain knowledge

Domain knowledge itself changes over time, and new KPIs and metrics can emerge within a company. For example, when social media marketing expanded rapidly, metrics that were not previously used suddenly gained importance. While multi-agent LLMs are flexible, problems arise regarding 'whether they can properly recognize new domain concepts' and 'whether they can appropriately update old domain information'.

If such mechanisms are not in place, there is a risk that the automated system will become obsolete.

Model transparency and explainability

Furthermore, explainability is particularly important for business applications. Since multi-agent LLMs have independent roles for each agent and can leave their decision-making processes as logs, it is thought that transparency is easier to increase compared to a single LLM.

However, it is not easy to grasp all the prompts and responses that occurred while multiple agents interacted. In industries with many regulations or cases where high accountability is required for management decisions, the extent to which internal model processes can be disclosed will be a challenge.

Computational costs and implementation hurdles

When putting a multi-agent LLM system into actual operation in a company, considerations for computational costs and infrastructure development are also essential. Because multiple agents perform inference in parallel or in stages, resource burdens may increase compared to inference with a single LLM. Additionally, integration with existing cloud services and data infrastructure is necessary to implement message passing between agents and reference external knowledge.

Furthermore, attention must be paid to practical operational perspectives, such as LLM version management and API usage costs. Especially in large companies, there are cases where the amount of data handled is enormous and security requirements are strict, making it impossible to use external APIs directly. In such cases, it may be necessary to devise ways to run multi-agent LLMs in an on-premises environment.

Future outlook

Advanced human collaboration

Moving forward, the process design for how human decision-makers and on-site staff provide feedback on the analysis results and visualizations autonomously generated by multi-agent LLMs will become increasingly important.

In particular, some companies define workflows such as 'automation will not proceed until the department head approves the model's output.' Additionally, it is essential to align these systems with on-site operational designs, such as by placing restrictions on LLMs running in the cloud while ensuring flexibility within the scope of internal rules and compliance requirements.

Integration with Semantic Layers

Data warehouses and data lakes often already have company-specific semantic layers defined. For example, if an item labeled 'SKU' is a company-specific parts management code, its meaning should be registered in standard data modeling tools or business glossaries.

If multi-agent LLMs can reference this semantic information, the accuracy of domain detection and concept extraction will improve dramatically. As implied in research papers, as mechanisms for LLMs to freely incorporate external knowledge sources become more established, more practical applications in the context of corporate data utilization will become possible.

Collaboration with Other Advanced Technologies

In the future, multi-agent LLMs may be combined with technologies such as the following:

The development toward multimodal analysis, especially incorporating visual and audio information, is particularly interesting; attempts to gain new insights, such as by analyzing call center conversation content or store surveillance camera footage and cross-referencing it with sales data, may progress.

Points to Consider When Actually Implementing

From Small-Scale Experiments to PoC

It is generally common to start with a small-scale Proof of Concept (PoC) for multi-agent LLM frameworks rather than introducing everything into the production environment at once. The following steps can be considered:

Assessing Costs and ROI

When considering this in an enterprise domain, the perspective of Return on Investment (ROI) is indispensable. Running large-scale LLMs in a multi-agent configuration often requires cloud resources, and the running costs are by no means low. However, since it is expected to reduce the number of analysis projects and the man-hours previously allocated to specialized analysts, it is important to calculate the estimates while balancing both factors.

Security and Privacy

In the case of financial institutions or the medical industry, which handle highly confidential data, there are cases where it is difficult to send data to an external cloud. Considerations for security and privacy protection are required, such as whether to deploy multi-agent LLMs in an on-premises environment or to input data into the model after anonymization or masking. Furthermore, since there is a possibility that personal information or sensitive attributes may be included in the generated insights, an appropriate monitoring system is required.

Summary

The Data-to-Dashboard framework using multi-agent LLMs is expected to bring a new stage of automation and insight to corporate data analysis. It is highly attractive because it enables multifaceted analysis that incorporates domain knowledge, rather than just 'Q&A' or 'template-based visualization,' and can consistently handle everything up to providing detailed explanations for the visualization results.

On the other hand, there are the following considerations when introducing it.

Based on these key points, the direction indicated by the framework introduced in this article will likely become a powerful option for raising the analytical capabilities of the entire organization. While expert knowledge in operational design and data governance is necessary for actual business application, the autonomous and comprehensive analysis brought about by multi-agent LLMs is expected to become the standard for data analysis in the near future.

Afterword

In this article, we explained cutting-edge research called "Data-to-Dashboard," which utilizes LLMs to automatically generate insightful dashboards from raw data. Enterprise analysis using multi-agent LLMs is a field with innovative potential.

The process of accurately incorporating domain knowledge, instantly accessing diverse metrics, and even automatically generating visualizations will lead traditional BI and analytics to a higher level.

If you found this article helpful, I would be encouraged if you could "like" or "follow." I will continue to share practical know-how and the latest trends, so I hope you will read them.

⚠️ Important: Less than 30% of people can fully utilize ChatGPT
Why? It is the difference in "questioning ability."
Actually, there is a scientific law to "how to frame questions" that extracts 10 times the value from AI. It is a method proven by Google's 20% rule and recommended by Amazon's Bezos. Everything is revealed in "The Strongest Skill of the AI Era: Curiosity Power."
Now that AI can return "answers" instantly, the turning point of the game lies in the "question." Please check out my book, "The Strongest Skill of the AI Era: Curiosity Power," which condenses that secret and 13 practical tools you can use in your work tomorrow.

"Why don't you convey your thoughts more clearly?"
A timeless masterpiece that has been read for over 25 years since its release as a "bible" for thinking methods. This book explains everything about the "Pyramid Principle," a thinking organization technique developed at the prestigious consulting firm McKinsey.
If you acquire the technique of thinking from the conclusion and structuring the evidence, your reports and presentations will become dramatically easier to understand. Here is a lifelong skill that allows you to grasp complex problems simply and dramatically increase your persuasiveness.
It is a must-read for all business people, regardless of era or occupation.

It identifies the essence of a question and systematizes the three steps of grasping the current situation -> extracting issues -> formulating solutions. It features a structure that trains "essential logical thinking" that can also be used for strategy firm interview preparation through examples and exercises. The latest work in the "Todai Note" series with over 350,000 copies sold.
Through practical case problems such as "How to increase convenience store sales?", it thoroughly trains not just knowledge or frameworks, but the "thinking process" itself that sees through the essence of things.
Perfect for those who want to acquire "thinking power" that can be used for a lifetime, rather than just knowledge. A must-read book for everyone from job seekers to veteran business people to take their thinking skills to the next level.

Why do highly educated people get scammed and experts make elementary mistakes? This book scientifically elucidates the shocking truth that "the higher the intelligence, the easier it is to fall into thinking traps."
The author, Robson, analyzes a wealth of cases from the failures of Nobel Prize winners to the misdiagnoses of doctors. He exposes the true nature of "cognitive bias" that cannot be prevented by IQ or academic background alone. The paradox that smart people are more likely to make big mistakes by overestimating their own knowledge and neglecting critical thinking.
What this book presents is not just a warning. It is packed with practical wisdom that is useful in all situations, from business decision-making and investment judgments to daily life, including specific methods for acquiring "metacognition" and "intellectual humility."
A book that should be read by those who think "I'm fine." By knowing the limits of intelligence, you will acquire true intelligence.

The definitive edition that serves as a compass for the generative AI era. The Matsuo-Iwasawa Laboratory at the University of Tokyo has gathered its full strength to thoroughly explain the latest trends in generative AI. Leading Japanese AI researchers comprehensively analyze the current state and future of generative AI technology, which has evolved rapidly since the advent of ChatGPT.
Published in March 2025, it comprehensively explains the information you need to know now, from the core of the technology to business applications, domestic and international laws and regulations, and safety. It also covers the latest trends such as AI agents and multimodal capabilities.
A must-have book for everyone involved in the generative AI business. It is a reliable source of information for making sound judgments in the rapidly changing AI industry.