Overcome Failing Document Ingestion & RAG Strategies with Agentic Knowledge Distillation

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:

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:

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!

Stay Ahead

Explore More Insights

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.

ExxonMobil begins drilling wells in Guyana’s EEZ

bp completes sale of German refinery, associated assets

European independent refiner Klesch Group has completed its previously announced deal to acquire bp plc’s 265,000-b/d refinery and related assets in Gelsenkirchen and Horst and Scholven, Germany, which is operated as an integrated refining and petrochemical site. With the transaction finalized as of Aug. 3, Klesch has taken full ownership

Samsung offers future AI memory roadmap

It is already being used now in NAND flash memory for 3D stacking. Rather than spread the memory circuits out, they are stacked on top of each other like stories on a high-rise building. The technique was first introduced in 2014, with 24-layer NAND flash period last year it broke

AMD to buy Taalas, maker of model-specific AI chips for enterprise inference

Where model-specific silicon fits Those tradeoffs significantly narrow the range of enterprise workloads where model-specific silicon is likely to make economic sense. Dai sees the technology as best suited for mature, predictable inference workloads that run at massive scale and rely on relatively stable AI models, such as customer service

TotalEnergies to acquire Shell’s European onshore renewables portfolio

TotalEnergies SE has agreed to acquire Shell’s 4-Gw onshore renewables portfolio in Europe. The portfolio includes 500 Mw of solar and wind assets in operation or under construction, primarily in Italy and the Netherlands, as well as a 3.5-Gw pipeline of solar, wind, and battery storage projects in Italy, the UK, and Spain, the company said Aug. 3. In the Netherlands, the assets include 254.2 Mw of installed peak capacity across the Moerdijk, Heerenveen-Zuid, and Emmen (GZI Next) solar parks; the Sas van Gent-Zuid and Koegorspolder solar parks in Terneuzen; and the Pottendijk combined solar and wind park in Emmen. TotalEnergies will assume full ownership of the portfolio upon closing. The transaction is subject to regulatory approvals and is expected to be completed by yearend 2026. “This agreement reflects Shell’s continued focus on actively managing and further strengthening its electricity portfolio, in line with the strategy outlined during Capital Markets Day 2025,” said Machteld de Haan, president, downstream, renewables and energy solutions, Shell. De Haan said Shell is prioritizing investment in areas where it has competitive advantages, including asset-backed power trading and customer-focused energy solutions. Shell said it will continue to buy and sell onshore solar and wind power in Europe and will retain interests in projects including Holland Hydrogen 1, Northern Lights CCS in Norway, LNG, and carbon capture and storage activities. KKR acquires 50% interest in European renewables portfolio In another deal, TotalEnergies agreed to farm out a 50% interest in a largely developed 1.2-Gw onshore solar and wind portfolio in Europe to KKR. The company said the transaction is consistent with its strategy of selling 50% interests in renewable assets once they have been developed. The portfolio includes assets in Germany, Spain, France, and Poland. Electricity generated by the assets has already been sold to third parties or

Petrobras makes another gas discovery offshore Colombia

@import url(‘https://fonts.googleapis.com/css2?family=Inter:wght@100..900&display=swap’); .ebm-page__main h1, .ebm-page__main h2, .ebm-page__main h3, .ebm-page__main h4, .ebm-page__main h5, .ebm-page__main h6 { font-family: Inter; } body { line-height: 150%; letter-spacing: 0.025em; } button, .ebm-button-wrapper { font-family: Inter; } .label-style { text-transform: uppercase; color: var(–color-grey); font-weight: 600; font-size: 0.75rem; } .caption-style { font-size: 0.75rem; opacity: .6; } #onetrust-pc-sdk [id*=btn-handler], #onetrust-pc-sdk [class*=btn-handler] { background-color: #c19a06 !important; border-color: #c19a06 !important; } #onetrust-policy a, #onetrust-pc-sdk a, #ot-pc-content a { color: #c19a06 !important; } #onetrust-consent-sdk #onetrust-pc-sdk .ot-active-menu { border-color: #c19a06 !important; } #onetrust-consent-sdk #onetrust-accept-btn-handler, #onetrust-banner-sdk #onetrust-reject-all-handler, #onetrust-consent-sdk #onetrust-pc-btn-handler.cookie-setting-link { background-color: #c19a06 !important; border-color: #c19a06 !important; } #onetrust-consent-sdk .onetrust-pc-btn-handler { color: #c19a06 !important; border-color: #c19a06 !important; } Petroleo Brasileiro SA (Petrobras) discovered another new gas accumulation with the deepwater Sandia-1 exploratory well in GUA-OFF-0 block about 42 km offshore Colombia in 1,251 m of water. Sandia-1 spudded on June 12, 2026, reaching final depth on July 29, 2026. Proven gas intervals are being evaluated by means of well profiles and will later be characterized by laboratory analyses. The well is 18 km from the Sirius-1 (discoverer) and Sirius-2 (appraisal) wells and 9 km from the Copoazu-1 (discoverer) well, indicating that there is strong gas potential in this area of offshore Colombia, Petrobras said. Petrobras subsidiary Petrobras International Braspetro BV–Sucursal Colombia (PIB-COL; 44.44%) operates the GUA-OFF-0 block on behalf of partner Ecopetrol SA (55.56%).

OPEC+ approves September output hike, completes 2023 cuts rollback

OPEC+ has approved a fresh increase in oil production quotas for September of roughly 188,000 b/d, completing the phased reversal of voluntary supply cuts first introduced in 2023. The decision, confirmed in an official OPEC statement following a virtual meeting on Aug. 2, 2026, marks the sixth consecutive monthly increase by the group this year. Seven core members of the alliance—Saudi Arabia, Russia, Iraq, Kuwait, Kazakhstan, Algeria, and Oman—agreed to raise output targets. The move completes the unwinding of the 1.65-million b/d voluntary supply cut originally agreed in 2023, back when the group still included the United Arab Emirates (UAE), which exited OPEC in May. The group said the adjustment would also give participating countries an opportunity to accelerate compensation for previous overproduction, and it reiterated commitment to the OPEC+ Declaration of Cooperation, with compliance to be monitored by the Joint Ministerial Monitoring Committee (JMMC). While the September hike is now finalized, OPEC+ is widely expected to pause further increases starting in the fourth quarter. Though the group’s official statement gave no explicit guidance on fourth-quarter policy, OPEC+ sources cited by Reuters and analysts—including Rystad Energy’s Jorge Leon—say a pause is likely as the alliance assesses market conditions after finishing the restoration of the 2023 cuts. A separate layer of roughly 2 million b/d in cuts, dating to 2022, remains in place and is expected to continue through the end of 2026. The steady stream of monthly increases comes against a backdrop of major market disruption. Ongoing Middle East tensions—including disruptions tied to the Iran conflict and the Strait of Hormuz—have complicated the group’s ability to translate higher quotas into actual barrels reaching the market. Russia, in particular, continues to produce below its OPEC+ target of about 9.8 million b/d, with output near 9 million b/d amid repeated Ukrainian drone

Market Focus: Reading the oil market after the US-Iran MOU collapse

Drawing on nearly three decades of experience in energy trading and risk management, Kessler offers insight into the fallout from escalating Middle East tensions, the breakdown of US-Iran diplomatic efforts, and the critical role of the Strait of Hormuz, through which a significant share of global oil supplies traditionally flows. The discussion explores what it would take to achieve a meaningful de-escalation in the region and how market participants are assessing the risks. Kessler argues that restoring safe passage through the Strait of Hormuz will be central to any lasting stability, while Iran’s oil exports and broader economic pressures could influence future negotiations. He also shares his perspective on how OPEC+ is responding to disruptions, the alliance’s efforts to restore production, and the growing competitive pressure it faces from producers outside the Gulf region. Turning to North America, Kessler examines the outlook for US shale producers in a higher-price environment. With crude prices holding above $80/bbl, he discusses signs of increased drilling activity, stronger production growth potential, and the continued emphasis on hedging and capital discipline among operators. The conversation also highlights advances in drilling technology and efficiency that could enable US producers to respond more quickly to market opportunities while managing downside risk. Looking further ahead, the episode considers whether recent disruptions will accelerate a long-term shift away from traditional Middle East oil chokepoints. Kessler discusses the growing role of US, Canadian, African, and Latin American supplies, expanding export infrastructure, and the possibility that today’s high prices could ultimately lead to demand destruction, increased competition, and renewed market oversupply. For anyone following global crude markets, OPEC+ strategy, US shale growth, energy security, and future oil price trends, this conversation provides a timely and thought-provoking outlook on the evolving global energy landscape. About our guest Dennis Kissler, senior vice-president of

Maurel & Prom to acquire Gran Tierra Energy’s assets in Colombia, Ecuador for $1.33 billion

The company said the predominantly operated portfolio comprises producing assets, development projects, and exploration acreage across Colombia’s Middle Magdalena Valley, Putumayo, and Llanos basins and Ecuador’s Oriente basin. Production is entirely oil-weighted and benefits from established processing, storage, and transportation infrastructure as well as access to multiple export routes. The principal Colombian assets include Acordionero, Costayaco, and Moqueta on the Chaza block, the Suroriente block centered on Cohembi, and the recently acquired interests in Tisquirama and San Roque. Growth opportunities in Colombia include continued development of Tisquirama, expansion of the Cohembi-Raju area, the Pegasus prospect, and longer-term potential associated with the La Luna formation. In Ecuador, the Chanangue, Charapa, Conejo, Iguana, Perico, and Espejo assets provide a combination of producing fields, discovered resources, and appraisal and exploration opportunities. Maurel & Prom said the assets represent a growth platform supported by existing discoveries and additional potential through waterflood application across the portfolio. For Gran Tierra Energy, the transaction serves as an exit from South America as part of the company’s plan to reduce debt and focus on growth opportunities in Canada and Azerbaijan. Maurel & Prom is a Paris-listed international oil and natural gas exploration and production company majority owned by PT Pertamina Internasional Eksplorasi dan Produksi (PIEP), a subsidiary of Indonesia’s national energy company, PT Pertamina (Persero). Closing, expected by yearend, is subject to shareholder approval, creditor consents, regulatory approvals, and other customary closing conditions.

HF Sinclair inks supply deals amid pending segment spinoff, refinery closure

HF Sinclair Corp. has lined up long-term supply arrangements to support the transition of the company’s lubricants and specialties products business in parallel with its recently announced plan to retire its 15,600-b/d base oil refining plant in Mississauga, Ont. After revealing a downstream integration strategy on July 28 involving the proposed closure of its Canadian refining business and transformation of its lubricants and specialties products segment, HF Sinclair confirmed on Aug. 3 that it entered into strategic long-term commercial agreements with suppliers SK On Co. Ltd.’s SK Enmove and Chevron USA Inc.’s Chevron Products Co. to establish a diversified North American base oil supply network. As part of the August agreement that aims to support maintaining base oil coverage after the Canadian refining assets are retired, SK Enmove and Chevron Products will supply HF Sinclair with Group III and Group II base oils, respectively, according to the companies. In return, HF Sinclair said its lubricants and specialties business will serve as a distributor for SK Enmove’s YUBASE Group III base oils in key regional markets in North America, as well as distribute Chevron-branded Group II base oils in Canada and select US regions. The supply arrangements come as part of HF Sinclair’s transformation and separation of its lubricants and specialties segments via the capital markets to create “two independent public companies in a manner that is tax-efficient for HF Sinclair and its shareholders,” according to the operator’s July 28 presentation to investors. HF Sinclair said it expects the new lubricants-specialties company will to drive organic growth and consolidate a highly fragmented global lubricants and specialties market, leading to reduced earnings volatility supported by diversified end markets and a differentiated finished and specialty mix of products. Subject to customary conditions and final approvals, HF Sinclair said the separation of the lubricants-specialty

Polish data center plans to send its waste heat to the neighbors

As Europe swelters in a heatwave, residents probably don’t want to hear about ways to make their homes even hotter, but that’s what Polish property developer Citylink is talking about, with plans to dump waste heat from a new data center in Wrocław into the municipal district heating network. Citylink is designing the data center so that heat from servers can be recovered instead of being dissipated via cooling systems — and as the data center grows, any increase in computing power will mean more energy available for recovery. The collaboration with local power company Kogeneracja will provide “valuable experience in designing and operating modern data centers, with a particular focus on infrastructure dedicated to AI nodes,” said Michał Starybrat, development director at Citylink.

The Data Center Industry’s Permission to Build

The data center industry has spent the past several years announcing the future. Gigawatts. AI factories. New regions. New power architectures. Campuses at a scale that would have seemed extraordinary before generative AI reset the industry’s expectations. Now the public has entered the room. Communities are asking harder questions about who pays for electrical infrastructure, where the water comes from, how much noise reaches neighboring properties and what remains locally after construction crews leave. Utilities are being pressed to protect ratepayers from speculative load and costly system upgrades. Elected officials who once treated data centers primarily as economic-development wins are finding that the politics have changed. The defining question is no longer whether demand is real. It is whether the data center industry can keep earning the permission required to build at the scale it has promised. I mean permission in a broader sense than zoning approval, an environmental permit or a signed utility agreement. I mean the political and social room to develop infrastructure measured in hundreds of megawatts and billions of dollars—often in places whose residents have only recently begun to understand what is being proposed around them. That room is narrowing. A Different Kind of Constraint On July 18, opponents organized 142 demonstrations across 42 states in what Reuters described as the first coordinated national protest against the data center buildout. The movement crossed familiar political boundaries, bringing together environmental advocates, rural landowners and residents concerned about power prices, water, noise and the pace of development. A June Reuters/Ipsos poll found that 57% of respondents would oppose a data center in their community. Only 14% said they would be comfortable with one nearby. Those findings deserve the industry’s full attention. New York has imposed a one-year pause on certain environmental approvals for new hyperscale data centers while

NVIDIA’s Reported $50B Lease and the Nuclear-Powered AI Factory

Aalo and Crusoe Pursue the Nuclear-Powered AI Factory The Aalo-Crusoe partnership addresses the industry’s power problem by bringing power generation directly to the compute. In this case, skipping intermediary power stages such as minimal grid or custom BTM gas turbine solutions and going straight to nuclear. Aalo Atomics and Crusoe said they plan to deploy a Crusoe Spark modular data center running Crusoe Cloud at Idaho National Laboratory in 2027. The proof-of-concept project is intended to demonstrate an AI workload operating on power from an Aalo advanced reactor. Crusoe continues to expand their other data center campus projects. The companies then intend to deploy Aalo Pods, Aalo’s 50-megawatt-electric nuclear power plants, at Crusoe data centers by the end of 2029. Aalo has already begun work on a second reactor beside its initial test unit at the Idaho site. That reactor is expected to produce electricity for the Crusoe installation. On July 4, 2026, Aalo’s zero-power Critical Test Reactor reached criticality, sustaining a nuclear chain reaction without generating commercial electricity. The test reactor contains a full-scale core and components analogous to those planned for the 10-megawatt-electric Aalo-X power reactor being built next door, but it operates before sodium coolant and electricity-generating systems are added. Aalo plans to continue experiments with the Critical Test Reactor to refine its reactor-physics models, characterize control behavior and generate data supporting development and licensing of the full-power Aalo-X system. Advanced nuclear announcements sometimes blur the line between a successful test, an electricity-producing demonstration and a commercially licensed fleet. Aalo has achieved an important technical milestone, but substantial work remains before reactors can be manufactured, licensed, financed and operated at commercial data center sites. The pairing with Crusoe should be noted because it connects a reactor developer with a company that can provide the data center load,

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 DesignNew York, NY (remote)This position is also available as a remote role anywhere in the US 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 CentersColumbus, OH (limited travel) Non-traveling CxA positions available in: Indianapolis, IN; Cedar Rapids, IA; Phoenix, AZ; Atlanta, GA and Austin, TX. Traveling CxA based really 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 and 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 leading-edge technology

Navigating Virginia’s Data Center Boom: Policy Shifts, Local Projects, and Future Challenges

Virginia’s newest high profile data center story is no longer the announcement of the next generation AI data center campus, it is now how the state is beginning to set the trend for legislative process to protect its communities while still encouraging the data center industry development. On August 3, state Senators Richard Stuart, a Republican, and Russet Perry, a Democrat, called on Gov. Abigail Spanberger to convene a special legislative session to address groundwater strain. Their request followed a state study warning that eastern Virginia’s groundwater supply is constrained and that large new industrial withdrawals may be difficult to sustain. The debate has expanded into calls for a broader pause: Senator Glen Sturtevant has asked for an immediate statewide moratorium on new data center development, while Senate President Pro Tempore Louise Lucas has said such a moratorium deserves serious consideration. Those proposals are not yet law, but they are the clearest indication that Virginia’s policy discussion has moved beyond incremental regulation. The Commonwealth spent years treating data centers primarily as an economic-development and tax-base success. It is now evaluating them simultaneously as power, water, land-use, air-quality and ratepayer issues. That shift is especially important for projects outside Northern Virginia, where developers are increasingly pursuing large sites in communities with less experience reviewing hyperscale infrastructure. The calls for a special session arrive only weeks after a significant package of data center laws and budget provisions took effect July 1. Virginia’s new budget established what the administration describes as a first-of-its-kind electricity consumption tax on data centers. The charge is 1.1 cents per kilowatt-hour, began July 1 and is capped at $600 million in annual collections, with excess revenue refunded to data center taxpayers. The compromise preserved Virginia’s sales-and-use-tax exemption for qualifying data center equipment, avoiding the abrupt repeal sought by

Land and Expand: The Gigawatt Credibility Test

The midsummer wave of U.S. data center development is not defined by a single market, developer or technology company. It stretches from the Georgia coast to West Texas, from the industrial Midwest to the Mississippi River. What links the projects announced since early June is not just their scale, it is the realization that scale alone is not enough. Developers are still announcing multibillion-dollar campuses and gigawatt power requirements, but the language surrounding those announcements has changed. Companies are emphasizing who will pay for new generation and transmission, how cooling systems will limit water consumption, what communities will receive beyond temporary construction employment, and when contracted customers will begin occupying capacity. In several cases, the announcement is less about acquiring land than proving that a project has become commercially and electrically credible. As we have seen progressing through the industry, the latest announcements point toward campuses that combine compute, power, financing and community agreements in one development package. OpenAI Goes Direct in Georgia OpenAI, on July 22 disclosed Project Camellia, a long-term data center development in Effingham County, Georgia. OpenAI said it is designing and developing the campus itself and has contracted with Georgia Power for 3.2 gigawatts of electricity, to be delivered in phases from 2028 through 2032. The project has been reported as a roughly $20 billion investment on approximately 1,400 acres, making it one of the largest individual data center proposals currently moving through the U.S. pipeline. Project Camellia is notable not only for its size but for OpenAI’s more direct role. The company has traditionally secured capacity through cloud providers and infrastructure partners. By taking responsibility for designing and developing the Georgia campus, OpenAI is signaling that control over power, schedule and facility design has become strategically important as AI companies compete for increasingly scarce large-scale

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.

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

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

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