Get ONMINE

Stay Ahead, Stay ONMINE

Overcome Failing Document Ingestion & RAG Strategies with Agentic Knowledge Distillation

Introduction Many generative AI use cases still revolve around Retrieval Augmented Generation (RAG), yet consistently fall short of user expectations. Despite the growing body of research on RAG improvements and even adding Agents into the process, many solutions still fail to return exhaustive results, miss information that is critical but infrequently mentioned in the documents, require multiple search iterations, and generally struggle to reconcile key themes across multiple documents. To top it all off, many implementations still rely on cramming as much “relevant” information as possible into the model’s context window alongside detailed system and user prompts. Reconciling all this information often exceeds the model’s cognitive capacity and compromises response quality and consistency. This is where our Agentic Knowledge Distillation + Pyramid Search Approach comes into play. Instead of chasing the best chunking strategy, retrieval algorithm, or inference-time reasoning method, my team, Jim Brown, Mason Sawtell, Sandi Besen, and I, take an agentic approach to document ingestion. We leverage the full capability of the model at ingestion time to focus exclusively on distilling and preserving the most meaningful information from the document dataset. This fundamentally simplifies the RAG process by allowing the model to direct its reasoning abilities toward addressing the user/system instructions rather than struggling to understand formatting and disparate information across document chunks.  We specifically target high-value questions that are often difficult to evaluate because they have multiple correct answers or solution paths. These cases are where traditional RAG solutions struggle most and existing RAG evaluation datasets are largely insufficient for testing this problem space. For our research implementation, we downloaded annual and quarterly reports from the last year for the 30 companies in the DOW Jones Industrial Average. These documents can be found through the SEC EDGAR website. The information on EDGAR is accessible and able to be downloaded for free or can be queried through EDGAR public searches. See the SEC privacy policy for additional details, information on the SEC website is “considered public information and may be copied or further distributed by users of the web site without the SEC’s permission”. We selected this dataset for two key reasons: first, it falls outside the knowledge cutoff for the models evaluated, ensuring that the models cannot respond to questions based on their knowledge from pre-training; second, it’s a close approximation for real-world business problems while allowing us to discuss and share our findings using publicly available data.  While typical RAG solutions excel at factual retrieval where the answer is easily identified in the document dataset (e.g., “When did Apple’s annual shareholder’s meeting occur?”), they struggle with nuanced questions that require a deeper understanding of concepts across documents (e.g., “Which of the DOW companies has the most promising AI strategy?”). Our Agentic Knowledge Distillation + Pyramid Search Approach addresses these types of questions with much greater success compared to other standard approaches we tested and overcomes limitations associated with using knowledge graphs in RAG systems.  In this article, we’ll cover how our knowledge distillation process works, key benefits of this approach, examples, and an open discussion on the best way to evaluate these types of systems where, in many cases, there is no singular “right” answer. Building the pyramid: How Agentic Knowledge Distillation works Image by author and team depicting pyramid structure for document ingestion. Robots meant to represent agents building the pyramid. Overview Our knowledge distillation process creates a multi-tiered pyramid of information from the raw source documents. Our approach is inspired by the pyramids used in deep learning computer vision-based tasks, which allow a model to analyze an image at multiple scales. We take the contents of the raw document, convert it to markdown, and distill the content into a list of atomic insights, related concepts, document abstracts, and general recollections/memories. During retrieval it’s possible to access any or all levels of the pyramid to respond to the user request.  How to distill documents and build the pyramid:  Convert documents to Markdown: Convert all raw source documents to Markdown. We’ve found models process markdown best for this task compared to other formats like JSON and it is more token efficient. We used Azure Document Intelligence to generate the markdown for each page of the document, but there are many other open-source libraries like MarkItDown which do the same thing. Our dataset included 331 documents and 16,601 pages.  Extract atomic insights from each page: We process documents using a two-page sliding window, which allows each page to be analyzed twice. This gives the agent the opportunity to correct any potential mistakes when processing the page initially. We instruct the model to create a numbered list of insights that grows as it processes the pages in the document. The agent can overwrite insights from the previous page if they were incorrect since it sees each page twice. We instruct the model to extract insights in simple sentences following the subject-verb-object (SVO) format and to write sentences as if English is the second language of the user. This significantly improves performance by encouraging clarity and precision. Rolling over each page multiple times and using the SVO format also solves the disambiguation problem, which is a huge challenge for knowledge graphs. The insight generation step is also particularly helpful for extracting information from tables since the model captures the facts from the table in clear, succinct sentences. Our dataset produced 216,931 total insights, about 13 insights per page and 655 insights per document. Distilling concepts from insights: From the detailed list of insights, we identify higher-level concepts that connect related information about the document. This step significantly reduces noise and redundant information in the document while preserving essential information and themes. Our dataset produced 14,824 total concepts, about 1 concept per page and 45 concepts per document.  Creating abstracts from concepts: Given the insights and concepts in the document, the LLM writes an abstract that appears both better than any abstract a human would write and more information-dense than any abstract present in the original document. The LLM generated abstract provides incredibly comprehensive knowledge about the document with a small token density that carries a significant amount of information. We produce one abstract per document, 331 total. Storing recollections/memories across documents: At the top of the pyramid we store critical information that is useful across all tasks. This can be information that the user shares about the task or information the agent learns about the dataset over time by researching and responding to tasks. For example, we can store the current 30 companies in the DOW as a recollection since this list is different from the 30 companies in the DOW at the time of the model’s knowledge cutoff. As we conduct more and more research tasks, we can continuously improve our recollections and maintain an audit trail of which documents these recollections originated from. For example, we can keep track of AI strategies across companies, where companies are making major investments, etc. These high-level connections are super important since they reveal relationships and information that are not apparent in a single page or document. Sample subset of insights extracted from IBM 10Q, Q3 2024 (page 4) We store the text and embeddings for each layer of the pyramid (pages and up) in Azure PostgreSQL. We originally used Azure AI Search, but switched to PostgreSQL for cost reasons. This required us to write our own hybrid search function since PostgreSQL doesn’t yet natively support this feature. This implementation would work with any vector database or vector index of your choosing. The key requirement is to store and efficiently retrieve both text and vector embeddings at any level of the pyramid.  This approach essentially creates the essence of a knowledge graph, but stores information in natural language, the way an LLM natively wants to interact with it, and is more efficient on token retrieval. We also let the LLM pick the terms used to categorize each level of the pyramid, this seemed to let the model decide for itself the best way to describe and differentiate between the information stored at each level. For example, the LLM preferred “insights” to “facts” as the label for the first level of distilled knowledge. Our goal in doing this was to better understand how an LLM thinks about the process by letting it decide how to store and group related information.  Using the pyramid: How it works with RAG & Agents At inference time, both traditional RAG and agentic approaches benefit from the pre-processed, distilled information ingested in our knowledge pyramid. The pyramid structure allows for efficient retrieval in both the traditional RAG case, where only the top X related pieces of information are retrieved or in the Agentic case, where the Agent iteratively plans, retrieves, and evaluates information before returning a final response.  The benefit of the pyramid approach is that information at any and all levels of the pyramid can be used during inference. For our implementation, we used PydanticAI to create a search agent that takes in the user request, generates search terms, explores ideas related to the request, and keeps track of information relevant to the request. Once the search agent determines there’s sufficient information to address the user request, the results are re-ranked and sent back to the LLM to generate a final reply. Our implementation allows a search agent to traverse the information in the pyramid as it gathers details about a concept/search term. This is similar to walking a knowledge graph, but in a way that’s more natural for the LLM since all the information in the pyramid is stored in natural language. Depending on the use case, the Agent could access information at all levels of the pyramid or only at specific levels (e.g. only retrieve information from the concepts). For our experiments, we did not retrieve raw page-level data since we wanted to focus on token efficiency and found the LLM-generated information for the insights, concepts, abstracts, and recollections was sufficient for completing our tasks. In theory, the Agent could also have access to the page data; this would provide additional opportunities for the agent to re-examine the original document text; however, it would also significantly increase the total tokens used.  Here is a high-level visualization of our Agentic approach to responding to user requests: Image created by author and team providing an overview of the agentic research & response process Results from the pyramid: Real-world examples To evaluate the effectiveness of our approach, we tested it against a variety of question categories, including typical fact-finding questions and complex cross-document research and analysis tasks.  Fact-finding (spear fishing):  These tasks require identifying specific information or facts that are buried in a document. These are the types of questions typical RAG solutions target but often require many searches and consume lots of tokens to answer correctly.  Example task: “What was IBM’s total revenue in the latest financial reporting?” Example response using pyramid approach: “IBM’s total revenue for the third quarter of 2024 was $14.968 billion [ibm-10q-q3-2024.pdf, pg. 4] Total tokens used to research and generate response This result is correct (human-validated) and was generated using only 9,994 total tokens, with 1,240 tokens in the generated final response.  Complex research and analysis:  These tasks involve researching and understanding multiple concepts to gain a broader understanding of the documents and make inferences and informed assumptions based on the gathered facts. Example task: “Analyze the investments Microsoft and NVIDIA are making in AI and how they are positioning themselves in the market. The report should be clearly formatted.” Example response: Response generated by the agent analyzing AI investments and positioning for Microsoft and NVIDIA. The result is a comprehensive report that executed quickly and contains detailed information about each of the companies. 26,802 total tokens were used to research and respond to the request with a significant percentage of them used for the final response (2,893 tokens or ~11%). These results were also reviewed by a human to verify their validity. Snippet indicating total token usage for the task Example task: “Create a report on analyzing the risks disclosed by the various financial companies in the DOW. Indicate which risks are shared and unique.” Example response: Part 1 of response generated by the agent on disclosed risks. Part 2 of response generated by the agent on disclosed risks. Similarly, this task was completed in 42.7 seconds and used 31,685 total tokens, with 3,116 tokens used to generate the final report.  Snippet indicating total token usage for the task These results for both fact-finding and complex analysis tasks demonstrate that the pyramid approach efficiently creates detailed reports with low latency using a minimal amount of tokens. The tokens used for the tasks carry dense meaning with little noise allowing for high-quality, thorough responses across tasks. Benefits of the pyramid: Why use it? Overall, we found that our pyramid approach provided a significant boost in response quality and overall performance for high-value questions.  Some of the key benefits we observed include:  Reduced model’s cognitive load: When the agent receives the user task, it retrieves pre-processed, distilled information rather than the raw, inconsistently formatted, disparate document chunks. This fundamentally improves the retrieval process since the model doesn’t waste its cognitive capacity on trying to break down the page/chunk text for the first time.  Superior table processing: By breaking down table information and storing it in concise but descriptive sentences, the pyramid approach makes it easier to retrieve relevant information at inference time through natural language queries. This was particularly important for our dataset since financial reports contain lots of critical information in tables.  Improved response quality to many types of requests: The pyramid enables more comprehensive context-aware responses to both precise, fact-finding questions and broad analysis based tasks that involve many themes across numerous documents.  Preservation of critical context: Since the distillation process identifies and keeps track of key facts, important information that might appear only once in the document is easier to maintain. For example, noting that all tables are represented in millions of dollars or in a particular currency. Traditional chunking methods often cause this type of information to slip through the cracks.  Optimized token usage, memory, and speed: By distilling information at ingestion time, we significantly reduce the number of tokens required during inference, are able to maximize the value of information put in the context window, and improve memory use.  Scalability: Many solutions struggle to perform as the size of the document dataset grows. This approach provides a much more efficient way to manage a large volume of text by only preserving critical information. This also allows for a more efficient use of the LLMs context window by only sending it useful, clear information. Efficient concept exploration: The pyramid enables the agent to explore related information similar to navigating a knowledge graph, but does not require ever generating or maintaining relationships in the graph. The agent can use natural language exclusively and keep track of important facts related to the concepts it’s exploring in a highly token-efficient and fluid way.  Emergent dataset understanding: An unexpected benefit of this approach emerged during our testing. When asking questions like “what can you tell me about this dataset?” or “what types of questions can I ask?”, the system is able to respond and suggest productive search topics because it has a more robust understanding of the dataset context by accessing higher levels in the pyramid like the abstracts and recollections.  Beyond the pyramid: Evaluation challenges & future directions Challenges While the results we’ve observed when using the pyramid search approach have been nothing short of amazing, finding ways to establish meaningful metrics to evaluate the entire system both at ingestion time and during information retrieval is challenging. Traditional RAG and Agent evaluation frameworks often fail to address nuanced questions and analytical responses where many different responses are valid. Our team plans to write a research paper on this approach in the future, and we are open to any thoughts and feedback from the community, especially when it comes to evaluation metrics. Many of the existing datasets we found were focused on evaluating RAG use cases within one document or precise information retrieval across multiple documents rather than robust concept and theme analysis across documents and domains.  The main use cases we are interested in relate to broader questions that are representative of how businesses actually want to interact with GenAI systems. For example, “tell me everything I need to know about customer X” or “how do the behaviors of Customer A and B differ? Which am I more likely to have a successful meeting with?”. These types of questions require a deep understanding of information across many sources. The answers to these questions typically require a person to synthesize data from multiple areas of the business and think critically about it. As a result, the answers to these questions are rarely written or saved anywhere which makes it impossible to simply store and retrieve them through a vector index in a typical RAG process.  Another consideration is that many real-world use cases involve dynamic datasets where documents are consistently being added, edited, and deleted. This makes it difficult to evaluate and track what a “correct” response is since the answer will evolve as the available information changes.  Future directions In the future, we believe that the pyramid approach can address some of these challenges by enabling more effective processing of dense documents and storing learned information as recollections. However, tracking and evaluating the validity of the recollections over time will be critical to the system’s overall success and remains a key focus area for our ongoing work.  When applying this approach to organizational data, the pyramid process could also be used to identify and assess discrepancies across areas of the business. For example, uploading all of a company’s sales pitch decks could surface where certain products or services are being positioned inconsistently. It could also be used to compare insights extracted from various line of business data to help understand if and where teams have developed conflicting understandings of topics or different priorities. This application goes beyond pure information retrieval use cases and would allow the pyramid to serve as an organizational alignment tool that helps identify divergences in messaging, terminology, and overall communication.  Conclusion: Key takeaways and why the pyramid approach matters The knowledge distillation pyramid approach is significant because it leverages the full power of the LLM at both ingestion and retrieval time. Our approach allows you to store dense information in fewer tokens which has the added benefit of reducing noise in the dataset at inference. Our approach also runs very quickly and is incredibly token efficient, we are able to generate responses within seconds, explore potentially hundreds of searches, and on average use

Introduction

Many generative AI use cases still revolve around Retrieval Augmented Generation (RAG), yet consistently fall short of user expectations. Despite the growing body of research on RAG improvements and even adding Agents into the process, many solutions still fail to return exhaustive results, miss information that is critical but infrequently mentioned in the documents, require multiple search iterations, and generally struggle to reconcile key themes across multiple documents. To top it all off, many implementations still rely on cramming as much “relevant” information as possible into the model’s context window alongside detailed system and user prompts. Reconciling all this information often exceeds the model’s cognitive capacity and compromises response quality and consistency.

This is where our Agentic Knowledge Distillation + Pyramid Search Approach comes into play. Instead of chasing the best chunking strategy, retrieval algorithm, or inference-time reasoning method, my team, Jim Brown, Mason Sawtell, Sandi Besen, and I, take an agentic approach to document ingestion.

We leverage the full capability of the model at ingestion time to focus exclusively on distilling and preserving the most meaningful information from the document dataset. This fundamentally simplifies the RAG process by allowing the model to direct its reasoning abilities toward addressing the user/system instructions rather than struggling to understand formatting and disparate information across document chunks. 

We specifically target high-value questions that are often difficult to evaluate because they have multiple correct answers or solution paths. These cases are where traditional RAG solutions struggle most and existing RAG evaluation datasets are largely insufficient for testing this problem space. For our research implementation, we downloaded annual and quarterly reports from the last year for the 30 companies in the DOW Jones Industrial Average. These documents can be found through the SEC EDGAR website. The information on EDGAR is accessible and able to be downloaded for free or can be queried through EDGAR public searches. See the SEC privacy policy for additional details, information on the SEC website is “considered public information and may be copied or further distributed by users of the web site without the SEC’s permission”. We selected this dataset for two key reasons: first, it falls outside the knowledge cutoff for the models evaluated, ensuring that the models cannot respond to questions based on their knowledge from pre-training; second, it’s a close approximation for real-world business problems while allowing us to discuss and share our findings using publicly available data. 

While typical RAG solutions excel at factual retrieval where the answer is easily identified in the document dataset (e.g., “When did Apple’s annual shareholder’s meeting occur?”), they struggle with nuanced questions that require a deeper understanding of concepts across documents (e.g., “Which of the DOW companies has the most promising AI strategy?”). Our Agentic Knowledge Distillation + Pyramid Search Approach addresses these types of questions with much greater success compared to other standard approaches we tested and overcomes limitations associated with using knowledge graphs in RAG systems. 

In this article, we’ll cover how our knowledge distillation process works, key benefits of this approach, examples, and an open discussion on the best way to evaluate these types of systems where, in many cases, there is no singular “right” answer.

Building the pyramid: How Agentic Knowledge Distillation works

AI-generated image showing a pyramid structure for document ingestion with labelled sections.
Image by author and team depicting pyramid structure for document ingestion. Robots meant to represent agents building the pyramid.

Overview

Our knowledge distillation process creates a multi-tiered pyramid of information from the raw source documents. Our approach is inspired by the pyramids used in deep learning computer vision-based tasks, which allow a model to analyze an image at multiple scales. We take the contents of the raw document, convert it to markdown, and distill the content into a list of atomic insights, related concepts, document abstracts, and general recollections/memories. During retrieval it’s possible to access any or all levels of the pyramid to respond to the user request. 

How to distill documents and build the pyramid: 

  1. Convert documents to Markdown: Convert all raw source documents to Markdown. We’ve found models process markdown best for this task compared to other formats like JSON and it is more token efficient. We used Azure Document Intelligence to generate the markdown for each page of the document, but there are many other open-source libraries like MarkItDown which do the same thing. Our dataset included 331 documents and 16,601 pages. 
  2. Extract atomic insights from each page: We process documents using a two-page sliding window, which allows each page to be analyzed twice. This gives the agent the opportunity to correct any potential mistakes when processing the page initially. We instruct the model to create a numbered list of insights that grows as it processes the pages in the document. The agent can overwrite insights from the previous page if they were incorrect since it sees each page twice. We instruct the model to extract insights in simple sentences following the subject-verb-object (SVO) format and to write sentences as if English is the second language of the user. This significantly improves performance by encouraging clarity and precision. Rolling over each page multiple times and using the SVO format also solves the disambiguation problem, which is a huge challenge for knowledge graphs. The insight generation step is also particularly helpful for extracting information from tables since the model captures the facts from the table in clear, succinct sentences. Our dataset produced 216,931 total insights, about 13 insights per page and 655 insights per document.
  3. Distilling concepts from insights: From the detailed list of insights, we identify higher-level concepts that connect related information about the document. This step significantly reduces noise and redundant information in the document while preserving essential information and themes. Our dataset produced 14,824 total concepts, about 1 concept per page and 45 concepts per document. 
  4. Creating abstracts from concepts: Given the insights and concepts in the document, the LLM writes an abstract that appears both better than any abstract a human would write and more information-dense than any abstract present in the original document. The LLM generated abstract provides incredibly comprehensive knowledge about the document with a small token density that carries a significant amount of information. We produce one abstract per document, 331 total.
  5. Storing recollections/memories across documents: At the top of the pyramid we store critical information that is useful across all tasks. This can be information that the user shares about the task or information the agent learns about the dataset over time by researching and responding to tasks. For example, we can store the current 30 companies in the DOW as a recollection since this list is different from the 30 companies in the DOW at the time of the model’s knowledge cutoff. As we conduct more and more research tasks, we can continuously improve our recollections and maintain an audit trail of which documents these recollections originated from. For example, we can keep track of AI strategies across companies, where companies are making major investments, etc. These high-level connections are super important since they reveal relationships and information that are not apparent in a single page or document.
Sample subset of insights extracted from IBM 10Q, Q3 2024
Sample subset of insights extracted from IBM 10Q, Q3 2024 (page 4)

We store the text and embeddings for each layer of the pyramid (pages and up) in Azure PostgreSQL. We originally used Azure AI Search, but switched to PostgreSQL for cost reasons. This required us to write our own hybrid search function since PostgreSQL doesn’t yet natively support this feature. This implementation would work with any vector database or vector index of your choosing. The key requirement is to store and efficiently retrieve both text and vector embeddings at any level of the pyramid. 

This approach essentially creates the essence of a knowledge graph, but stores information in natural language, the way an LLM natively wants to interact with it, and is more efficient on token retrieval. We also let the LLM pick the terms used to categorize each level of the pyramid, this seemed to let the model decide for itself the best way to describe and differentiate between the information stored at each level. For example, the LLM preferred “insights” to “facts” as the label for the first level of distilled knowledge. Our goal in doing this was to better understand how an LLM thinks about the process by letting it decide how to store and group related information. 

Using the pyramid: How it works with RAG & Agents

At inference time, both traditional RAG and agentic approaches benefit from the pre-processed, distilled information ingested in our knowledge pyramid. The pyramid structure allows for efficient retrieval in both the traditional RAG case, where only the top X related pieces of information are retrieved or in the Agentic case, where the Agent iteratively plans, retrieves, and evaluates information before returning a final response. 

The benefit of the pyramid approach is that information at any and all levels of the pyramid can be used during inference. For our implementation, we used PydanticAI to create a search agent that takes in the user request, generates search terms, explores ideas related to the request, and keeps track of information relevant to the request. Once the search agent determines there’s sufficient information to address the user request, the results are re-ranked and sent back to the LLM to generate a final reply. Our implementation allows a search agent to traverse the information in the pyramid as it gathers details about a concept/search term. This is similar to walking a knowledge graph, but in a way that’s more natural for the LLM since all the information in the pyramid is stored in natural language.

Depending on the use case, the Agent could access information at all levels of the pyramid or only at specific levels (e.g. only retrieve information from the concepts). For our experiments, we did not retrieve raw page-level data since we wanted to focus on token efficiency and found the LLM-generated information for the insights, concepts, abstracts, and recollections was sufficient for completing our tasks. In theory, the Agent could also have access to the page data; this would provide additional opportunities for the agent to re-examine the original document text; however, it would also significantly increase the total tokens used. 

Here is a high-level visualization of our Agentic approach to responding to user requests:

Overview of the agentic research & response process
Image created by author and team providing an overview of the agentic research & response process

Results from the pyramid: Real-world examples

To evaluate the effectiveness of our approach, we tested it against a variety of question categories, including typical fact-finding questions and complex cross-document research and analysis tasks. 

Fact-finding (spear fishing): 

These tasks require identifying specific information or facts that are buried in a document. These are the types of questions typical RAG solutions target but often require many searches and consume lots of tokens to answer correctly. 

Example task: “What was IBM’s total revenue in the latest financial reporting?”

Example response using pyramid approach: “IBM’s total revenue for the third quarter of 2024 was $14.968 billion [ibm-10q-q3-2024.pdf, pg. 4]

Screenshot of total tokens used to research and generate response
Total tokens used to research and generate response

This result is correct (human-validated) and was generated using only 9,994 total tokens, with 1,240 tokens in the generated final response. 

Complex research and analysis: 

These tasks involve researching and understanding multiple concepts to gain a broader understanding of the documents and make inferences and informed assumptions based on the gathered facts.

Example task: “Analyze the investments Microsoft and NVIDIA are making in AI and how they are positioning themselves in the market. The report should be clearly formatted.”

Example response:

Screenshot of the response generated by the agent analyzing AI investments and positioning for Microsoft and NVIDIA.
Response generated by the agent analyzing AI investments and positioning for Microsoft and NVIDIA.

The result is a comprehensive report that executed quickly and contains detailed information about each of the companies. 26,802 total tokens were used to research and respond to the request with a significant percentage of them used for the final response (2,893 tokens or ~11%). These results were also reviewed by a human to verify their validity.

Screenshot of snippet indicating total token usage for the task
Snippet indicating total token usage for the task

Example task: “Create a report on analyzing the risks disclosed by the various financial companies in the DOW. Indicate which risks are shared and unique.”

Example response:

Screenshot of part 1 of a response generated by the agent on disclosed risks.
Part 1 of response generated by the agent on disclosed risks.
Screenshot of part 2 of a response generated by the agent on disclosed risks.
Part 2 of response generated by the agent on disclosed risks.

Similarly, this task was completed in 42.7 seconds and used 31,685 total tokens, with 3,116 tokens used to generate the final report. 

Screenshot of a snippet indicating total token usage for the task
Snippet indicating total token usage for the task

These results for both fact-finding and complex analysis tasks demonstrate that the pyramid approach efficiently creates detailed reports with low latency using a minimal amount of tokens. The tokens used for the tasks carry dense meaning with little noise allowing for high-quality, thorough responses across tasks.

Benefits of the pyramid: Why use it?

Overall, we found that our pyramid approach provided a significant boost in response quality and overall performance for high-value questions. 

Some of the key benefits we observed include: 

  • Reduced model’s cognitive load: When the agent receives the user task, it retrieves pre-processed, distilled information rather than the raw, inconsistently formatted, disparate document chunks. This fundamentally improves the retrieval process since the model doesn’t waste its cognitive capacity on trying to break down the page/chunk text for the first time. 
  • Superior table processing: By breaking down table information and storing it in concise but descriptive sentences, the pyramid approach makes it easier to retrieve relevant information at inference time through natural language queries. This was particularly important for our dataset since financial reports contain lots of critical information in tables. 
  • Improved response quality to many types of requests: The pyramid enables more comprehensive context-aware responses to both precise, fact-finding questions and broad analysis based tasks that involve many themes across numerous documents. 
  • Preservation of critical context: Since the distillation process identifies and keeps track of key facts, important information that might appear only once in the document is easier to maintain. For example, noting that all tables are represented in millions of dollars or in a particular currency. Traditional chunking methods often cause this type of information to slip through the cracks. 
  • Optimized token usage, memory, and speed: By distilling information at ingestion time, we significantly reduce the number of tokens required during inference, are able to maximize the value of information put in the context window, and improve memory use. 
  • Scalability: Many solutions struggle to perform as the size of the document dataset grows. This approach provides a much more efficient way to manage a large volume of text by only preserving critical information. This also allows for a more efficient use of the LLMs context window by only sending it useful, clear information.
  • Efficient concept exploration: The pyramid enables the agent to explore related information similar to navigating a knowledge graph, but does not require ever generating or maintaining relationships in the graph. The agent can use natural language exclusively and keep track of important facts related to the concepts it’s exploring in a highly token-efficient and fluid way. 
  • Emergent dataset understanding: An unexpected benefit of this approach emerged during our testing. When asking questions like “what can you tell me about this dataset?” or “what types of questions can I ask?”, the system is able to respond and suggest productive search topics because it has a more robust understanding of the dataset context by accessing higher levels in the pyramid like the abstracts and recollections. 

Beyond the pyramid: Evaluation challenges & future directions

Challenges

While the results we’ve observed when using the pyramid search approach have been nothing short of amazing, finding ways to establish meaningful metrics to evaluate the entire system both at ingestion time and during information retrieval is challenging. Traditional RAG and Agent evaluation frameworks often fail to address nuanced questions and analytical responses where many different responses are valid.

Our team plans to write a research paper on this approach in the future, and we are open to any thoughts and feedback from the community, especially when it comes to evaluation metrics. Many of the existing datasets we found were focused on evaluating RAG use cases within one document or precise information retrieval across multiple documents rather than robust concept and theme analysis across documents and domains. 

The main use cases we are interested in relate to broader questions that are representative of how businesses actually want to interact with GenAI systems. For example, “tell me everything I need to know about customer X” or “how do the behaviors of Customer A and B differ? Which am I more likely to have a successful meeting with?”. These types of questions require a deep understanding of information across many sources. The answers to these questions typically require a person to synthesize data from multiple areas of the business and think critically about it. As a result, the answers to these questions are rarely written or saved anywhere which makes it impossible to simply store and retrieve them through a vector index in a typical RAG process. 

Another consideration is that many real-world use cases involve dynamic datasets where documents are consistently being added, edited, and deleted. This makes it difficult to evaluate and track what a “correct” response is since the answer will evolve as the available information changes. 

Future directions

In the future, we believe that the pyramid approach can address some of these challenges by enabling more effective processing of dense documents and storing learned information as recollections. However, tracking and evaluating the validity of the recollections over time will be critical to the system’s overall success and remains a key focus area for our ongoing work. 

When applying this approach to organizational data, the pyramid process could also be used to identify and assess discrepancies across areas of the business. For example, uploading all of a company’s sales pitch decks could surface where certain products or services are being positioned inconsistently. It could also be used to compare insights extracted from various line of business data to help understand if and where teams have developed conflicting understandings of topics or different priorities. This application goes beyond pure information retrieval use cases and would allow the pyramid to serve as an organizational alignment tool that helps identify divergences in messaging, terminology, and overall communication. 

Conclusion: Key takeaways and why the pyramid approach matters

The knowledge distillation pyramid approach is significant because it leverages the full power of the LLM at both ingestion and retrieval time. Our approach allows you to store dense information in fewer tokens which has the added benefit of reducing noise in the dataset at inference. Our approach also runs very quickly and is incredibly token efficient, we are able to generate responses within seconds, explore potentially hundreds of searches, and on average use (this includes all the search iterations!). 

We find that the LLM is much better at writing atomic insights as sentences and that these insights effectively distill information from both text-based and tabular data. This distilled information written in natural language is very easy for the LLM to understand and navigate at inference since it does not have to expend unnecessary energy reasoning about and breaking down document formatting or filtering through noise. 

The ability to retrieve and aggregate information at any level of the pyramid also provides significant flexibility to address a variety of query types. This approach offers promising performance for large datasets and enables high-value use cases that require nuanced information retrieval and analysis. 

Note: The opinions expressed in this article are solely my own and do not necessarily reflect the views or policies of my employer.

Interested in discussing further or collaborating? Reach out on LinkedIn!

Shape
Shape
Stay Ahead

Explore More Insights

More Articles

Stay ahead with more perspectives on cutting-edge power, infrastructure, energy,  bitcoin and AI solutions. Explore these articles to uncover strategies and insights shaping the future of industries.

Shape

AI data boom gives tape storage a new lease on life

“We are seeing unprecedented data growth combined with increasing cost, energy, and cyber resilience pressures across the industry,” said Hugues Meyrath, CEO of Quantum in a statement. “As organizations adapt to this new reality, tape is increasingly viewed as a strategic component of modern data infrastructure, delivering predictable economics and

Read More »

DOE’s Alternative Fuels and Feedstocks Office Announces up to $58 Million to Promote Chemical Innovation

WASHINGTON—The U.S. Department of Energy’s (DOE) Alternative Fuels and Feedstocks Office (AFFO) today announced up to $58 million in funding to advance novel, high-impact chemical technologies that use domestically sourced alternative and waste feedstocks. Projects funded through this initiative will advance new methods of chemical production that maximize the use of America’s vast biomass and waste resources. This funding supports President Trump’s Executive Order, Unleashing American Energy, which calls for targeted federal investment in technology innovation that strengthens the U.S. chemical sector.  “By investing in projects that use our abundant domestic resources and build strong industry partnerships, DOE will bolster American chemical manufacturing,” said AFFO Director Valerie Sarisky-Reed. “This funding will turn cutting-edge research into market-ready industrial solutions, strengthening our chemical supply chain, lowering costs for U.S. businesses and consumers, and securing America’s economic future.” The Accelerating Scale-up and Pre-piloting of Emerging Chemical Technologies (ASPECT) funding opportunity promotes the development and commercialization of chemical technologies that lower costs, enhance performance, reduce reliance on imports, and unlock strong market growth potential. ASPECT seeks to reduce time to market by moving projects from laboratory research to pre-pilot scale testing. It includes two main topic areas:  Topic Area 1: Bench ASPECT Proposals should support the development and adoption of new technologies for producing chemicals from alternative feedstocks, moving beyond proof-of-concept to bench and pre-pilot scale. Topic Area 2: Pre-pilot ASPECT Proposals should aim to accelerate the development and market entry of strategically valuable, domestically produced chemicals. AFFO will host an informational webinar for potential applicants on September 11, 2026, to explain the streamlined application and review process.  Applicants must submit concept papers by October 9, 2026, at 5:00 p.m. ET, to be eligible to submit a Stage 1 full application. To learn more about topic areas, registration requirements, applicant eligibility, webinar registration, and the Teaming Partner list, visit the

Read More »

Energy Secretary Secures Carolinas’ Grid Ahead of Holiday Weekend

WASHINGTON—The U.S. Department of Energy (DOE) today issued an emergency order to mitigate the risk of blackouts in the Carolinas amid hot weather conditions. Issued pursuant to Section 202(c) of the Federal Power Act, the order authorizes Duke Energy Carolinas, LLC (Duke) to dispatch specified units and to order their operation as needed to maintain reliability. The order also authorizes Duke, in collaboration with its Transmission Owners, to direct backup generation resources to operate as a last resort before declaring an Energy Emergency Alert (EEA) 3 or during an EEA 3. This order was issued pursuant to an application from Duke submitted on September 3, 2026. “Thanks to this emergency order, Americans will not have to worry about losing access to affordable power this Labor Day weekend,” said U.S. Secretary of Energy Chris Wright. “The previous administration’s energy subtraction policies weakened the grid, leaving Americans more vulnerable during events like this. Under President Trump’s leadership, we are ensuring that hardworking American families and businesses in the Carolinas’ have continued access energy to power and cool their homes.” On day one, President Trump declared a national energy emergency after the Biden administration’s energy subtraction agenda left behind a grid increasingly vulnerable to the risk of blackouts. The order is in effect beginning on September 3, 2026, through September 8, 2026. 

Read More »

President Trump’s Energy Dominance Agenda is Delivering for American Energy Workers

WASHINGTON—This Labor Day, the U.S. Department of Energy (DOE) is celebrating the hardworking men and women who power America with the release of the 2026 U.S. Energy and Employment Report (USEER). The annual report highlights strong job growth across critical energy sectors at the heart of President Trump’s Energy Dominance agenda. America’s most reliable energy sectors are adding jobs and powering industries across the country. These critical sectors deliver the affordable, reliable, and secure energy that American families, businesses, and industries depend on. After years of decline under the previous administration, America’s coal and nuclear power workforces are growing again under President Trump’s leadership. “Energy is the sector that enables every other sector of our economy, and America’s energy workers make it all possible,” said U.S. Secretary of Energy Chris Wright. “These hardworking men and women keep our lights on, our factories running, and our economy growing. President Trump’s Energy Dominance agenda is putting them first and delivering the affordable, reliable, and secure energy America needs.” Energy careers are also delivering bigger paychecks for American workers. The median energy-sector salary reached $63,000—24% higher than the U.S. median salary. America’s growing energy needs are creating the jobs of the future. The 2026 USEER’s new Future Outlook chapter highlights rising demand for skilled energy workers and growing competition for talent across energy and other expanding industries. These trends are opening new pathways to high-paying, skilled careers for American workers. As energy demand grows, America’s energy workforce will power the next generation of American industry, innovation, and economic growth. Highlights from the report include:  •    The median energy-sector salary was $63,000, 24% higher than the national median salary of $51,000. •    Natural gas transmission and distribution added 12,500 workers, growing employment by 5%. •    Nuclear power added 2,300 workers, growing employment by 4%. •    Coal power generation added 2,800 workers, growing

Read More »

Hydrocarbons and Geothermal Energy Office Issues Request for Information to Advance Private Investment in Innovative American Energy Technologies

WASHINGTON — The U.S. Department of Energy’s (DOE) Hydrocarbons and Geothermal Energy Office (HGEO), in collaboration with the Office of Technology Commercialization, today announced a Request for Information (RFI) seeking stakeholder input on opportunities to better connect private capital and public-private partnership programs with coal, oil and gas, and geothermal energy technologies.  The RFI supports President Trump’s American Energy Dominance agenda by strengthening connections among American energy innovators, industry and private capital to accelerate commercialization, lower energy costs, strengthen reliability and energy security, and power American prosperity. This effort builds on the recently announced Small Business Investment Company-Energy (SBIC-E) Initiative, announced by U.S. Energy Secretary Chris Wright and SBA Administrator Kelly Loeffler to mobilize private capital for American energy technologies and businesses.  “President Trump has made American energy innovation and dominance a priority, and connecting promising technologies with the right technical expertise, industry partners and sources of capital is critical to delivering on that vision,” said DOE Acting Assistant Secretary for the Hydrocarbons and Geothermal Energy Office Curt Coccodrilli. “Through this Request for Information, we would like to hear directly from investors, innovators and industry about the opportunities and challenges they see in commercializing coal, oil and gas, and geothermal energy technologies.” “Too often, promising American technologies face barriers between development and commercial deployment,” said Anthony Pugliese, DOE Chief Commercialization Officer and Director of the Office of Technology Commercialization. “We want to better understand where those barriers exist and how DOE can work with the private sector to create stronger pathways to market, helping more American energy technologies scale, compete, and succeed.”  DOE is soliciting feedback from investors, industry, academia, research laboratories, government agencies and other stakeholders to better understand investor interest, barriers and opportunities related to the development and commercialization of subsurface energy technologies.   Areas of interest include, but are not

Read More »

Energy Department Announces Geothermal Center of Excellence to Advance Geothermal Technology Innovation and Development

WASHINGTON — The U.S. Department of Energy’s (DOE) Hydrocarbons and Geothermal Energy Office today established a Geothermal Center of Excellence (CoE) to advance President Trump and Secretary Wright’s commitment to delivering affordable, reliable, and secure energy.  The Center will unite expertise and world-class capabilities from across DOE’s National Laboratories to accelerate the discovery and development of gigawatt-scale geothermal energy, resource discovery, and commercial development.  The National Laboratory of the Rockies (NLR) will lead the consortium, with support from the National Energy Technology Laboratory (NETL).  “America has vast geothermal resources beneath our feet that can provide reliable, around-the-clock energy while strengthening our energy dominance,” said DOE Under Secretary for Energy Kyle Haustveit. “Under President Trump’s leadership, the Geothermal Center of Excellence will leverage the world-class scientific and engineering expertise of our national laboratories in partnership with industry to unlock gigawatt-scale power generation, expand American energy production, and deliver more affordable, reliable, and secure energy to the American people.” DOE formally launched the Center at NLR’s campus in Golden, Colorado, bringing together DOE leadership, elected officials, NLR and NETL leadership, laboratory staff, and industry representatives to advance the Center’s vision and priorities.  “The Geothermal Center of Excellence marks an important step in our work to accelerate gigawatt-scale geothermal energy on the U.S. grid,” said DOE Acting Assistant Secretary for the Hydrocarbons and Geothermal Energy Office Curt Coccodrilli. “By driving innovation and enhancing lab-industry collaboration, the Center will help us achieve our goals to enhance reliable baseload power, strengthen grid reliability, and improve long-term energy security.” The Center will also serve as industry’s main entry point to DOE’s National Laboratories. An Industry Advisory Board will provide objective insight into industry-relevant geothermal research needs, accelerate industry-lab partnerships, and advise on Center priorities.  For more information, contact geo.centerofexcellence@nlr.gov.

Read More »

San Matías Pipeline secures $900 million for Vaca Muerta-to-LNG gas pipeline

The remaining $400 million will be contributed by the consortium’s shareholders: Pan American Energy, YPF, Pampa Energía, Harbour Energy, and Golar LNG. The 472-km, 36-in. OD San Matías Pipeline, which will originate at Tratayén, one of Vaca Muerta’s main gas hubs, is designed to transport 27 million cu m/d (MMcmd) of natural gas, aligned with the gas requirements of the two FLNG units. Hilli Episeyo will have LNG production capacity of 2.45 million tonnes/year (tpy) and will require about 11.5 MMcmd of feed gas. Esperanza, previously known as MKII, will add another 3.5 million tpy and require close to 16 MMcmd of feed gas. Hilli Episeyo is expected to begin operations in 2027, followed by Esperanza in 2028. The project also will include a compressor station with about 46,000 hp of installed capacity to maintain required pressure and flow across the system. Pipeline construction has been awarded to the SICIM-Víctor Contreras consortium, while OPS will be responsible for the Allen compressor station. IEB Construcciones was selected to manage and coordinate the project’s different construction fronts. In August, the first 36-in. line pipe manufactured in India began arriving at the Port of San Antonio Este. Construction is scheduled to begin in August 2026, with completion targeted for mid-2028. The project was admitted to Argentina’s Large Investment Incentive Regime (RIGI) in June and has environmental impact approvals from Neuquén and Río Negro provinces.

Read More »

ERCOT Puts Texas AI Megawatts to the Test

Texas has no shortage of proposed data center megawatts. The harder question is how many of them are real. That distinction is becoming central to the Electric Reliability Council of Texas (ERCOT) as the state works through an unprecedented wave of AI, hyperscale and other large-load requests. In June, ERCOT said it was tracking more than 438 GW of proposed large loads, nearly 89% associated with data centers. By Aug. 3, Gov. Greg Abbott said ERCOT was considering approximately 474 GW of connection requests, roughly 90% from data centers and more than five times the system’s record peak demand. Neither figure represents a forecast of what will actually get built. And that is increasingly the point. ERCOT’s new Batch Zero process is beginning to put harder boundaries around Texas’ enormous development pipeline, asking which projects have enough maturity, technical information and commitment to warrant space in the transmission plan. At the same time, new requirements surrounding voltage ride-through and dynamic modeling are forcing another realization on the AI infrastructure industry: at hundreds of megawatts, a data center is no longer simply a customer at the edge of the grid. Its behavior can affect the grid itself. For developers, utilities and investors, Texas is becoming a large-scale test of what separates an announced AI campus from executable infrastructure. The Queue Is Not the Grid The sheer scale of ERCOT’s large-load queue can obscure how early many projects remain. ERCOT’s April 2026 monthly report offered a revealing snapshot. Large-load applications totaled 445.8 GW through 2033, but 321 GW had no studies submitted to ERCOT. Another 93.7 GW was under ERCOT review, while 22 GW had met the applicable Section 9.5 requirements. Against that enormous development funnel, ERCOT reported just 5.9 GW of observed energized large loads, with another 3.2 GW approved to

Read More »

DCF Trends Summit: AI Compresses the Data Center Hardware Lifecycle and Raises the Stakes for ITAD

The AI infrastructure race is largely a story about getting more computing into data centers faster. But the accelerated hardware cycle is creating an equally consequential problem at the other end of the rack: getting yesterday’s equipment back out while it is still valuable. GPU systems built around increasingly dense and specialized AI architectures are beginning to challenge traditional assumptions about IT asset disposition, or ITAD. Where conventional enterprise infrastructure might remain in service for three to five years, newer GPU platforms can face refresh cycles of 18 to 24 months, according to Josh Humm, Data Center Solutions Manager at Dynamic Lifecycle Innovations. That compression changes the economics as well as the mechanics of decommissioning. “The faster we can get the materials out of your building, the more it’s worth, the more we can return to your program,” Humm said. Humm joined DCF Contributing Editor Doug Black for a DCF Show podcast recorded at the third annual Data Center Frontier Trends Summit, held Aug. 4-6 in Reston, Virginia. Their conversation focused on a less visible part of the AI infrastructure buildout: what happens to servers, accelerators, memory, storage and networking gear when the next generation arrives. The answer increasingly touches facility operations, data security, logistics, sustainability and potentially millions of dollars in recoverable hardware value. AI Hardware Changes the Exit Path AI systems create some obvious physical challenges for decommissioning. Traditional ITAD teams accustomed to pulling 1U and 2U servers out of air-cooled racks may instead encounter liquid-cooling manifolds, substantially heavier systems and equipment requiring specialized rigging and handling procedures. Humm said some systems can weigh between 5,000 and 6,000 pounds. “We’re not pulling out just 1U, 2U servers out of racks anymore,” he said. Liquid cooling adds another layer. Removing infrastructure designed around direct-to-chip or other liquid-cooling architectures can

Read More »

Data Center Jobs: Engineering, Construction, Commissioning, Sales, Field Service and Facility Tech Jobs Available in Major Data Center Hotspots

Each month Data Center Frontier, in partnership with Pkaza, posts some of the hottest data center career opportunities in the market. Here’s a look at some of the latest data center jobs posted on the Data Center Frontier jobs board, powered by Pkaza Critical Facilities Recruiting. Looking for Data Center Candidates? Check out Pkaza’s Active Candidate / Featured Candidate Hotlist  CFD Engineer – Data Center Mechanical Design New York, NY (remote)This position is also available as a remote role anywhere in the U.S. in addition to key markets such as Cedar Rapids, IA; Kansas City, CA or White Plains, NY. Our client is an engineering design and commissioning company that has a national footprint and specializes in MEP critical facilities design. They provide design, commissioning, consulting and management expertise in the critical facilities space. They have a mindset to provide reliability, energy efficiency, and sustainable design expertise when providing these consulting services for enterprise, colocation and hyperscale companies. This career-growth minded opportunity offers exciting projects with leading-edge technology and innovation as well as competitive salaries and benefits.  Electrical Commissioning Agent – Data Centers Columbus, OH (limited travel) Non-traveling CxA positions available in: Indianapolis, IN; Cedar Rapids, IA; Phoenix, AZ; Atlanta, GA and Austin, TX. Traveling CxA based near any major airport, otherwise traveling to: New York, NY; White Plains, NY; Dallas, TX; Richmond, VA; Montvale, NJ; Charlotte, NC; Salt Lake City, UT; Kansas City, MO; Chesterton, IN or Chicago, IL. *** Also looking for a lead EE, ME CxA agents and CxA PMs. *** This opportunity is with a leading EPC company of data center design / build / commissioning solutions. This company provides a complete life cycle of solutions that are custom-fit to the requirements of their client’s mission-critical facilities. This opportunity provides a career-growth minded role with exciting projects with

Read More »

DCFTS 2026: Data Center Development Moves From Projection to Execution

The Data Center Map Gets More Selective For EdgeCore, finding viable development locations has become an exercise in aggressive filtering. Kestler said the company evaluated 172 sites during the previous 12 months to narrow the field to seven locations it wanted to actively manage. Its requirements include roughly 100 acres or more, the ability to support a 300-MVA-or-larger substation, credible utility development timelines and sufficient network proximity to support what Kestler called “interdependent compute” locations. The distinction matters. Not every AI workload needs the same geography, and not every site marketed as available for AI infrastructure can support the combination of land, network, power and timing required to make a project real. Miller placed that process in the context of a data center map already being redrawn by power availability. Northern Virginia’s power constraints in 2022 provided an early warning, redirecting capacity into markets including Atlanta and driving developers farther afield in search of large blocks of electricity. AI has intensified the process. As campus requirements move toward hundreds of megawatts and, in some cases, gigawatt scale, Miller said, fewer locations can satisfy all of the requirements simultaneously. Community acceptance is narrowing the map further. At the same time, Miller pointed to a potential countertrend: the growth of AI inference could create another layer of data center geography. Some inference architectures may favor smaller, distributed facilities rather than concentrating every workload inside enormous campuses. The result could be a more stratified infrastructure market. “Everything everywhere all at once,” Miller said. Build Where Data Centers Are Wanted For large campus development, Kestler offered another increasingly important filter. EdgeCore wants to build where it is wanted. In practical terms, that means targeting municipalities and jurisdictions that have already made deliberate decisions about where data center or other light industrial development belongs. Kestler

Read More »

How States Are Rewriting the Rules for Data Center Growth

Pennsylvania has moved from courting data center investment to setting much stricter terms for how the industry grows. Governor Josh Shapiro’s August 18 executive order creates one of the country’s most comprehensive state-level frameworks for large data centers, linking a more favorable environmental permitting process and state tax treatment to requirements covering power supply, grid costs, local approval, workforce commitments, water use and environmental performance. The order is the latest stage of Shapiro’s Governor’s Responsible Infrastructure Development, or GRID, initiative. GRID was announced in February, detailed in May and partially reinforced through Pennsylvania’s 2026-27 budget in July. The Pennsylvania House also passed legislation intended to codify the standards, but the Senate did not act. Shapiro has now used existing executive and agency authority to put much of the framework into effect immediately. Pennsylvania’s debate has also produced more direct proposals to slow development. Senate Bill 1359 would impose a statewide moratorium on hyperscale data center development and permitting, although the measure remains in the Senate Local Government Committee. A separate measure, Senate Bill 1345, would authorize municipalities to temporarily stop accepting or considering new applications for high-impact data centers for up to 18 months. SB 1345 advanced to second consideration in the Senate in July. Neither measure has become law. What is the Impact on Data Center Development? For data center projects with peak demand exceeding 25 MW, Pennsylvania’s template GRID Consent Order and Agreement provides the mechanism for binding developers to the requirements while allowing the states Department of Environmental Protection (DEP) to review qualifying permit applications on a rolling basis. Developers that decline to sign can still seek permits, but DEP will not begin reviewing their applications until local approvals and required water or wastewater authorizations are secured, and permits will not be handled on a rolling basis.

Read More »

PwC Maps $31.6 Trillion AI Data Center Buildout Through 2050

The scale of the AI infrastructure buildout is becoming easier to describe in trillions than billions. PwC’s inaugural Global Data Centre Outlook 2026–50 projects $31.6 trillion in cumulative global data center capital expenditure through 2050 under its central scenario, with annual spending rising from roughly $800 billion in 2026 to $1.1 trillion in 2030 and $1.8 trillion by 2050. There is also an enormous range around that central case. PwC, working with Oxford Economics, puts plausible cumulative investment at roughly $22 trillion to nearly $50 trillion, depending primarily on how quickly AI adoption progresses. But the most important finding may not be the $31.6 trillion headline. PwC argues that the economics of AI infrastructure are creating a fundamentally different capital cycle from previous infrastructure booms. Data centers are long-lived assets, but the increasingly expensive computing equipment inside them is not. Servers, GPUs, networking systems and other information and communications technology equipment are expected to require replacement on roughly four- to six-year cycles. PwC calculates that every $1 of construction spending can effectively commit the market to approximately $12 of subsequent ICT investment. ICT equipment accounts for about 70% of total data center CapEx in 2026 under its model, rising to 93% by 2050. That creates something closer to a continuously renewing technology platform than a conventional construction cycle. Over a 20-year data center asset life, PwC estimates that a facility could undergo three to five rounds of ICT investment. Increasing rack densities can force corresponding power and cooling upgrades, but the largest recurring expense remains the compute hardware itself. For data center developers and operators, that distinction matters. The economic life of the building increasingly diverges from the technical and financial life of the infrastructure filling it. AI Fragments the Data Center Demand Model The report also sees AI broadening

Read More »

Microsoft will invest $80B in AI data centers in fiscal 2025

And Microsoft isn’t the only one that is ramping up its investments into AI-enabled data centers. Rival cloud service providers are all investing in either upgrading or opening new data centers to capture a larger chunk of business from developers and users of large language models (LLMs).  In a report published in October 2024, Bloomberg Intelligence estimated that demand for generative AI would push Microsoft, AWS, Google, Oracle, Meta, and Apple would between them devote $200 billion to capex in 2025, up from $110 billion in 2023. Microsoft is one of the biggest spenders, followed closely by Google and AWS, Bloomberg Intelligence said. Its estimate of Microsoft’s capital spending on AI, at $62.4 billion for calendar 2025, is lower than Smith’s claim that the company will invest $80 billion in the fiscal year to June 30, 2025. Both figures, though, are way higher than Microsoft’s 2020 capital expenditure of “just” $17.6 billion. The majority of the increased spending is tied to cloud services and the expansion of AI infrastructure needed to provide compute capacity for OpenAI workloads. Separately, last October Amazon CEO Andy Jassy said his company planned total capex spend of $75 billion in 2024 and even more in 2025, with much of it going to AWS, its cloud computing division.

Read More »

John Deere unveils more autonomous farm machines to address skill labor shortage

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More Self-driving tractors might be the path to self-driving cars. John Deere has revealed a new line of autonomous machines and tech across agriculture, construction and commercial landscaping. The Moline, Illinois-based John Deere has been in business for 187 years, yet it’s been a regular as a non-tech company showing off technology at the big tech trade show in Las Vegas and is back at CES 2025 with more autonomous tractors and other vehicles. This is not something we usually cover, but John Deere has a lot of data that is interesting in the big picture of tech. The message from the company is that there aren’t enough skilled farm laborers to do the work that its customers need. It’s been a challenge for most of the last two decades, said Jahmy Hindman, CTO at John Deere, in a briefing. Much of the tech will come this fall and after that. He noted that the average farmer in the U.S. is over 58 and works 12 to 18 hours a day to grow food for us. And he said the American Farm Bureau Federation estimates there are roughly 2.4 million farm jobs that need to be filled annually; and the agricultural work force continues to shrink. (This is my hint to the anti-immigration crowd). John Deere’s autonomous 9RX Tractor. Farmers can oversee it using an app. While each of these industries experiences their own set of challenges, a commonality across all is skilled labor availability. In construction, about 80% percent of contractors struggle to find skilled labor. And in commercial landscaping, 86% of landscaping business owners can’t find labor to fill open positions, he said. “They have to figure out how to do

Read More »

2025 playbook for enterprise AI success, from agents to evals

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More 2025 is poised to be a pivotal year for enterprise AI. The past year has seen rapid innovation, and this year will see the same. This has made it more critical than ever to revisit your AI strategy to stay competitive and create value for your customers. From scaling AI agents to optimizing costs, here are the five critical areas enterprises should prioritize for their AI strategy this year. 1. Agents: the next generation of automation AI agents are no longer theoretical. In 2025, they’re indispensable tools for enterprises looking to streamline operations and enhance customer interactions. Unlike traditional software, agents powered by large language models (LLMs) can make nuanced decisions, navigate complex multi-step tasks, and integrate seamlessly with tools and APIs. At the start of 2024, agents were not ready for prime time, making frustrating mistakes like hallucinating URLs. They started getting better as frontier large language models themselves improved. “Let me put it this way,” said Sam Witteveen, cofounder of Red Dragon, a company that develops agents for companies, and that recently reviewed the 48 agents it built last year. “Interestingly, the ones that we built at the start of the year, a lot of those worked way better at the end of the year just because the models got better.” Witteveen shared this in the video podcast we filmed to discuss these five big trends in detail. Models are getting better and hallucinating less, and they’re also being trained to do agentic tasks. Another feature that the model providers are researching is a way to use the LLM as a judge, and as models get cheaper (something we’ll cover below), companies can use three or more models to

Read More »

OpenAI’s red teaming innovations define new essentials for security leaders in the AI era

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More OpenAI has taken a more aggressive approach to red teaming than its AI competitors, demonstrating its security teams’ advanced capabilities in two areas: multi-step reinforcement and external red teaming. OpenAI recently released two papers that set a new competitive standard for improving the quality, reliability and safety of AI models in these two techniques and more. The first paper, “OpenAI’s Approach to External Red Teaming for AI Models and Systems,” reports that specialized teams outside the company have proven effective in uncovering vulnerabilities that might otherwise have made it into a released model because in-house testing techniques may have missed them. In the second paper, “Diverse and Effective Red Teaming with Auto-Generated Rewards and Multi-Step Reinforcement Learning,” OpenAI introduces an automated framework that relies on iterative reinforcement learning to generate a broad spectrum of novel, wide-ranging attacks. Going all-in on red teaming pays practical, competitive dividends It’s encouraging to see competitive intensity in red teaming growing among AI companies. When Anthropic released its AI red team guidelines in June of last year, it joined AI providers including Google, Microsoft, Nvidia, OpenAI, and even the U.S.’s National Institute of Standards and Technology (NIST), which all had released red teaming frameworks. Investing heavily in red teaming yields tangible benefits for security leaders in any organization. OpenAI’s paper on external red teaming provides a detailed analysis of how the company strives to create specialized external teams that include cybersecurity and subject matter experts. The goal is to see if knowledgeable external teams can defeat models’ security perimeters and find gaps in their security, biases and controls that prompt-based testing couldn’t find. What makes OpenAI’s recent papers noteworthy is how well they define using human-in-the-middle

Read More »
More Articles