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 agent traffic drives first profitable year for Fastly

Fetcher bots, which retrieve content in real time when users make queries to AI assistants, show different concentration patterns. OpenAI’s ChatGPT and related bots generated 68% of fetcher bot requests. In some cases, fetcher bot request volumes exceeded 39,000 requests per minute to individual sites. AI agents check multiple websites

Read More »

Naftogaz Seeks USA Funds to Renovate Destroyed Plants

Ukraine’s state-run oil and gas company Naftogaz Group is seeking funds to restore and renovate its facilities after the destruction caused by constant Russian attacks, said its top executive. At least €3 billion ($3.5 billion) of damage has been done to the country’s facilities, with equipment needs exceeding €900 million, according to the company.  Naftogaz is particularly interested in Ukraine’s ongoing talks with partners such as the US Exim Bank and the US International Development Finance Corp., Chief Executive Officer Sergii Koretskyi told Bloomberg News in an interview at his office in Kyiv. He also stressed the importance of European assistance. Some $250 million in unspent Ukraine assistance funds remain with the US State Department, he said — part of which could be used to purchase US-made gas compressor units to allow Kyiv to repair production facilities. Their use would also be a boon to American companies, he added.  “Now we need funding for imports, investments and technologies. This is definitely a win-win situation for all parties — we’re not saying ‘help us’ but offering mutually beneficial cooperation,” said Koretskyi. Naftogaz, which provides gas to 12.5 million households, is a key element of Ukraine’s energy sector. Its infrastructure, as well as that of other power companies, has come under intense Russian bombardment in recent weeks, depriving many civilians of heating amid freezing temperatures. Since the start of this year, Naftogaz infrastructure has already faced 20 strikes, damaging oil and gas production and transportation systems, Koretskyi said.  He said that last year was the most destructive for Ukraine’s energy sector since Russian President Vladimir Putin began his full-scale invasion nearly four years ago, with hundreds of missiles and drones hitting facilities. Last February and October were the hardest months for Naftogaz specifically, the CEO added. The company’s biggest challenge is the unpredictable consequences

Read More »

VEN Plans to Grant More Oil Blocks to Chevron and Repsol

Venezuela plans to grant more oil-production land to Chevron Corp. and Spain’s Repsol SA as the Trump administration pushes for private companies to rebuild the nation’s energy sector, according to people with knowledge of the matter. Officials in Caracas are poised to award the exploration and production blocks as soon as this week, the people said. Giving US and European companies more access to Venezuela’s oil-rich territory is a key piece of US President Donald Trump’s push to revive the nation’s dilapidated energy sector while eroding China and Russia’s local influence.  On Thursday, US Energy Secretary Chris Wright toured a project operated by Chevron in Venezuela’s Orinoco oil belt and told reporters that the opportunity for cooperation between the US and the South American nation is immense following the capture of former Venezuela President Nicolás Maduro.  In an interview with Bloomberg TV, Wright said the US would release additional licenses “soon,” with companies like Chevron seeing benefits from an increase of as much as a 30% in production in the next 18 to 24 months.  “Chevron is being enabled to massively grow their business here. They’re the largest producer in Venezuela today, and they’re going to be able to both expand the reserves they have and expand their operations,” Wright said. “They’re just one of many, but they’re going to be a big one,” he added. Repsol declined to comment. Chevron didn’t immediately respond to a request for comment. The Trump administration is expected to issue general license to allow international oil companies to explore and produce in Venezuela without violating US sanctions, Bloomberg reported earlier this month. It would be the latest is part of a string of authorizations from the Treasury Department to open up the nation’s oil sector since US forces captured Venezuela’s former President Nicolás Maduro on Jan.

Read More »

Oil Posts Second Straight Weekly Drop

Oil notched its first back-to-back weekly drop this year as traders weighed the prospect of expanded OPEC+ supplies against US-Iran nuclear talks and recent weakness in wider markets. West Texas Intermediate fell 1% for the week and ended the day little changed on Friday. President Donald Trump said the US deployed an additional aircraft carrier to the Middle East in case a nuclear deal is not reached with Iran. “If we don’t have a deal, we’ll need it,” Trump said at the White House. He added he thinks negotiations will ultimately be successful. Traders have been watching for any uptick in tensions between Washington and Tehran that could pose a threat to supply from the Middle East. The commodity was down earlier as OPEC+ members see scope for output increases to resume in April, believing concerns about a glut are overblown, delegates said. The group has not yet committed to any course of action or begun formal discussions for a March 1 meeting, they added. A second weekly decline in the futures market stands to snap a long run of gains for early 2026, when recurrent bouts of geopolitical tension including the US stand-off with Iran supported oil prices. At an energy conference in London this week, attendees flagged that they expect worldwide supplies to top demand this year, potentially feeding into higher inventories in the Atlantic basin, the region where global prices are set. Still, a pile-up of sanctioned oil coupled with supply disruptions in various nations has limited the impact thus far. Trading may be thinner ahead of the Presidents’ Day holiday in the US, contributing to exaggerated price swings. Oil Prices WTI for March delivery settled up 0.1% at $62.89 a barrel in New York. Brent for April settlement edged 0.3% higher to $67.75 a barrel. What

Read More »

Reliance Gets USA License to Directly Buy VEN Crude

Indian refiner Reliance Industries Ltd. has received a general license from the US government that will allow it to purchase Venezuelan oil directly, according to a person familiar with the matter.  Reliance, owned by billionaire Mukesh Ambani, applied for the permit last month and received it from the Treasury Department a few days ago, the person said, asking not to be named as the matter is not public. The move comes immediately on the heels of a trade deal with the US that slashes punitive tariffs for Indian exports but demands that the country stop importing discounted Russian oil. The Indian government has asked state-owned refiners to consider buying more Venezuelan crude, as well as oil from the US.  Venezuela is unlikely to produce large volumes of crude anytime soon, but even limited supplies provide a fallback option for India’s largest refiner. The US — which has stepped up involvement in Venezuela’s oil sector after capturing the country’s president last month — has been considering general licenses to permit purchases, trading and investment in a sprawling but threadbare industry. Reliance is the first Indian refiner to receive clearance in the current push.  Reliance has historically been an important consumer of the country’s heavy crude, having struck a term deal to secure as much as 400,000 barrels a day from Petroleos de Venezuela SA in 2012. It is among only a handful of refiners in India that have the capacity to process the high-viscosity, sour oil, which is difficult to extract and refine without diluent.  The Indian refining giant took about 25% of Venezuela’s exports in 2019, before its term deal got suspended in 2019 due to US sanctions. It last received a general license in 2024 and took crude until that expired last year, and was not renewed. Reuters first reported the issuance of

Read More »

Baker Hughes Explores $1.5B Sale of Waygate Unit

Baker Hughes Co. is exploring a potential sale of its Waygate Technologies unit, which provides industrial testing and inspection equipment, people with knowledge of the matter said.  The world’s second-biggest oilfield contractor is working with advisers to study a possible divestment of the Waygate business, which could fetch around $1.5 billion, according to the people. A sale process could kick off in the next few months and attract interest from private equity firms, the people said, asking not to be identified because the information is private.  Deliberations are ongoing and there’s no certainty they will lead to a transaction, the people said. A representative for Baker Hughes declined to comment.  Waygate, based in Hürth, Germany, makes radiographic testing systems, industrial CT scanners, remote visual inspection machines and ultrasonic testing devices. It operates in more than 80 countries and is known for brands including Krautkrämer, phoenix|x-ray, Seifert, Everest and Agfa NDT.  The company was started in 2004 as GE Inspection Technologies. It’s been under the current ownership since 2017, when General Electric Co. combined its oil and gas division with Baker Hughes in a $32 billion deal.  Baker Hughes is selling the non-core asset after agreeing last year to buy industrial equipment maker Chart Industries Inc. for about $9.6 billion in one of its biggest-ever acquisitions. Chief Executive Officer Lorenzo Simonelli said in October last year that Baker Hughes is undertaking a “comprehensive evaluation” of its capital allocation focus following the Chart deal in order to boost shareholder value.  The pending sale would join other sizeable corporate divestments in Europe. Volkswagen AG has launched the sale of a majority stake in its heavy diesel engine maker Everllence, while Continental AG is selling its Contitech business. WHAT DO YOU THINK? Generated by readers, the comments included herein do not reflect the views and opinions

Read More »

EIA Raises 2026 WTI Forecast, Lowers 2027 Projection

The U.S. Energy Information Administration (EIA) increased its 2026 West Texas Intermediate (WTI) crude oil average spot price forecast, and lowered its 2027 projection, in its latest short term energy outlook (STEO). According to the EIA’s February STEO, which was released on February 10, the EIA now sees the WTI spot price averaging $53.42 per barrel this year and $49.34 per barrel next year. In its previous STEO, which was released in January, the EIA projected that the WTI spot price would average $52.21 per barrel in 2026 and $50.36 per barrel in 2027. A quarterly breakdown included in the EIA’s latest STEO projected that the WTI average spot price will come in at $58.62 per barrel in the first quarter of this year, $53.65 per barrel in the second quarter, $51.69 per barrel in the third quarter, $50.00 per barrel in the fourth quarter, $49.00 per barrel in the first quarter of next year, $49.66 per barrel in the second quarter, $49.68 per barrel in the third quarter, and $49.00 per barrel in the fourth quarter of 2027. In its previous STEO, the EIA forecast that the WTI spot price would average $54.93 per barrel in the first quarter of this year, $52.67 per barrel in the second quarter, $52.03 per barrel in the third quarter, $49.34 per barrel in the fourth quarter, $49.00 per barrel in the first quarter of next year, $50.66 per barrel in the second quarter, $50.68 per barrel in the third quarter, and $51.00 per barrel in the fourth quarter of 2027. In a BMI report sent to Rigzone by the Fitch Group on Friday, BMI projected that the front month WTI crude price will average $64.00 per barrel in 2026 and $68.00 per barrel in 2027. Standard Chartered sees the NYMEX WTI nearby

Read More »

Arista laments ‘horrendous’ memory situation

Digging in on campus Arista has been clear about its plans to grow its presence campus networking environments. Last Fall, Ullal said she expects Arista’s campus and WAN business would grow from the current $750 million-$800 million run rate to $1.25 billion, representing a 60% growth opportunity for the company. “We are committed to our aggressive goal of $1.25 billion for ’26 for the cognitive campus and branch. We have also successfully deployed in many routing edge, core spine and peering use cases,” Ullal said. “In Q4 2025, Arista launched our flagship 7800 R4 spine for many routing use cases, including DCI, AI spines with that massive 460 terabits of capacity to meet the demanding needs of multiservice routing, AI workloads and switching use cases. The combined campus and routing adjacencies together contribute approximately 18% of revenue.” Ethernet leads the way “In terms of annual 2025 product lines, our core cloud, AI and data center products built upon our highly differentiated Arista EOS stack is successfully deployed across 10 gig to 800 gigabit Ethernet speeds with 1.6 terabit migration imminent,” Ullal said. “This includes our portfolio of EtherLink AI and our 7000 series platforms for best-in-class performance, power efficiency, high availability, automation, agility for both the front and back-end compute, storage and all of the interconnect zones.” Ullal said she expects Ethernet will get even more of a boost later this year when the multivendor Ethernet for Scale-Up Networking (ESUN) specification is released.  “We have consistently described that today’s configurations are mostly a combination of scale out and scale up were largely based on 800G and smaller ratings. Now that the ESUN specification is well underway, we need a good solid spec. Otherwise, we’ll be shipping proprietary products like some people in the world do today. And so we will tie our

Read More »

From NIMBY to YIMBY: A Playbook for Data Center Community Acceptance

Across many conversations at the start of this year, at PTC and other conferences alike, the word on everyone’s lips seems to be “community.” For the data center industry, that single word now captures a turning point from just a few short years ago: we are no longer a niche, back‑of‑house utility, but a front‑page presence in local politics, school board budgets, and town hall debates. That visibility is forcing a choice in how we tell our story—either accept a permanent NIMBY-reactive framework, or actively build a YIMBY narrative that portrays the real value digital infrastructure brings to the markets and surrounding communities that host it. Speaking regularly with Ilissa Miller, CEO of iMiller Public Relations about this topic, there is work to be done across the ecosystem to build communications. Miller recently reflected: “What we’re seeing in communities isn’t a rejection of digital infrastructure, it’s a rejection of uncertainty driven by anxiety and fear. Most local leaders have never been given a framework to evaluate digital infrastructure developments the way they evaluate roads, water systems, or industrial parks. When there’s no shared planning language, ‘no’ becomes the safest answer.” A Brief History of “No” Community pushback against data centers is no longer episodic; it has become organized, media‑savvy, and politically influential in key markets. In Northern Virginia, resident groups and environmental organizations have mobilized against large‑scale campuses, pressing counties like Loudoun and Prince William to tighten zoning, question incentives, and delay or reshape projects.1 Loudoun County’s move in 2025 to end by‑right approvals for new facilities, requiring public hearings and board votes, marked a watershed moment as the world’s densest data center market signaled that communities now expect more say over where and how these campuses are built. Prince William County’s decision to sharply increase its tax rate on

Read More »

Nomads at the Frontier: PTC 2026 Signals the Digital Infrastructure Industry’s Moment of Execution

Each January, the Pacific Telecommunications Council conference serves as a barometer for where digital infrastructure is headed next. And according to Nomad Futurist founders Nabeel Mahmood and Phillip Koblence, the message from PTC 2026 was unmistakable: The industry has moved beyond hype. The hard work has begun. In the latest episode of The DCF Show Podcast, part of our ongoing ‘Nomads at the Frontier’ series, Mahmood and Koblence joined Data Center Frontier to unpack the tone shift emerging across the AI and data center ecosystem. Attendance continues to grow year over year. Conversations remain energetic. But the character of those conversations has changed. As Mahmood put it: “The hype that the market started to see is actually resulting a bit more into actions now, and those conversations are resulting into some good progress.” The difference from prior years? Less speculation. More execution. From Data Center Cowboys to Real Deployments Koblence offered perhaps the sharpest contrast between PTC conversations in 2024 and those in 2026. Two years ago, many projects felt speculative. Today, developers are arriving with secured power, customers, and construction underway. “If 2024’s PTC was data center cowboys — sites that in someone’s mind could be a data center — this year was: show me the money, show me the power, give me accurate timelines.” In other words, the market is no longer rewarding hypothetical capacity. It is demanding delivered capacity. Operators now speak in terms of deployments already underway, not aspirational campuses still waiting on permits and power commitments. And behind nearly every conversation sits the same gating factor. Power. Power Has Become the Industry’s Defining Constraint Whether discussions centered on AI factories, investment capital, or campus expansion, Mahmood and Koblence noted that every conversation eventually returned to energy availability. “All of those questions are power,” Koblence said.

Read More »

Cooling Consolidation Hits AI Scale: LiquidStack, Submer, and the Future of Data Center Thermal Strategy

As AI infrastructure scales toward ever-higher rack densities and gigawatt-class campuses, cooling has moved from a technical subsystem to a defining strategic issue for the data center industry. A trio of announcements in early February highlights how rapidly the cooling and AI infrastructure stack is consolidating and evolving: Trane Technologies’ acquisition of LiquidStack; Submer’s acquisition of Radian Arc, extending its reach from core data centers into telco edge environments; and Submer’s partnership with Anant Raj to accelerate sovereign AI infrastructure deployment across India. Layered atop these developments is fresh guidance from Oracle Cloud Infrastructure explaining why closed-loop, direct-to-chip cooling is becoming central to next-generation facility design, particularly in regions where water use has become a flashpoint in community discussions around data center growth. Taken together, these developments show how the industry is moving beyond point solutions toward integrated, scalable AI infrastructure ecosystems, where cooling, compute, and deployment models must work together across hyperscale campuses and distributed edge environments alike. Trane Moves to Own the Cooling Stack The most consequential development comes from Trane Technologies, which on February 10 announced it has entered into a definitive agreement to acquire LiquidStack, one of the pioneers and leading innovators in data center liquid cooling. The acquisition significantly strengthens Trane’s ambition to become a full-service thermal partner for data center operators, extending its reach from plant-level systems all the way down to the chip itself. LiquidStack, headquartered in Carrollton, Texas, built its reputation on immersion cooling and advanced direct-to-chip liquid solutions supporting high-density deployments across hyperscale, enterprise, colocation, edge, and blockchain environments. Under Trane, those technologies will now be scaled globally and integrated into a broader thermal portfolio. In practical terms, Trane is positioning itself to deliver cooling across the full thermal chain, including: • Central plant equipment and chillers.• Heat rejection and controls

Read More »

Infrastructure Maturity Defines the Next Phase of AI Deployment

The State of Data Infrastructure Global Report 2025 from Hitachi Vantara arrives at a moment when the data center industry is undergoing one of the most profound structural shifts in its history. The transition from enterprise IT to AI-first infrastructure has moved from aspiration to inevitability, forcing operators, developers, and investors to confront uncomfortable truths about readiness, resilience, and risk. Although framed around “AI readiness,” the report ultimately tells an infrastructure story: one that maps directly onto how data centers are designed, operated, secured, and justified economically. Drawing on a global survey of more than 1,200 IT leaders, the report introduces a proprietary maturity model that evaluates organizations across six dimensions: scalability, reliability, security, governance, sovereignty, and sustainability. Respondents are then grouped into three categories—Emerging, Defined, and Optimized—revealing a stark conclusion: most organizations are not constrained by access to AI models or capital, but by the fragility of the infrastructure supporting their data pipelines. For the data center industry, the implications are immediate, shaping everything from availability design and automation strategies to sustainability planning and evolving customer expectations. In short, extracting value from AI now depends less on experimentation and more on the strength and resilience of the underlying infrastructure. The Focus of the Survey: Infrastructure, Not Algorithms Although the report is positioned as a study of AI readiness, its primary focus is not models, training approaches, or application development, but rather the infrastructure foundations required to operate AI reliably at scale. Drawing on responses from more than 1,200 organizations, Hitachi Vantara evaluates how enterprises are positioned to support production AI workloads across six dimensions as stated above: scalability, reliability, security, governance, sovereignty, and sustainability. These factors closely reflect the operational realities shaping modern data center design and management. The survey’s central argument is that AI success is no longer

Read More »

AI’s New Land Grab: Meta’s Indiana Megaproject and the Rise of Europe’s Neocloud Challengers

While Meta’s Indiana campus anchors hyperscale expansion in the United States, Europe recorded its own major infrastructure milestone this week as Amsterdam-based AI infrastructure provider Nebius unveiled plans for a 240-megawatt data center campus in Béthune, France, near Lille in the country’s northern industrial corridor. When completed, the campus will rank among Europe’s largest AI-focused data center facilities and positions northern France as a growing node in the continent’s expanding AI infrastructure map. The development repurposes a former Bridgestone tire manufacturing site, reflecting a broader trend across Europe in which legacy industrial properties, already equipped with heavy power access, transport links, and industrial zoning, are being converted into large-scale digital infrastructure hubs. Located within reach of connectivity and enterprise corridors linking Paris, Brussels, London, and Amsterdam, the site allows Nebius to serve major European markets while avoiding the congestion and power constraints increasingly shaping Tier 1 data center hubs. Industrial Infrastructure Becomes Digital Infrastructure Developers increasingly view former industrial sites as ideal for AI campuses because they often provide: • Existing grid interconnection capacity built for heavy industry• Transport and logistics infrastructure already in place• Industrial zoning that reduces permitting friction• Large contiguous parcels suited to phased campus expansion For regions like Hauts-de-France, redevelopment projects also offer economic transition opportunities, replacing legacy manufacturing capacity with next-generation digital infrastructure investment. Local officials have positioned the project as part of broader efforts to reposition northern France as a logistics and technology hub within Europe. The Neocloud Model Gains Ground Beyond the site itself, Nebius’ expansion illustrates the rapid emergence of neocloud infrastructure providers, companies building GPU-intensive AI capacity without operating full hyperscale cloud ecosystems. These firms increasingly occupy a strategic middle ground: supplying AI compute capacity to enterprises, startups, and even hyperscalers facing short-term infrastructure constraints. Nebius’ rise over the past year

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