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.

Microsoft’s largest quantum site to be built in Denmark

With this strategic move, Denmark will become Microsoft’s global quantum hub. According to the company, the expansion of the Lyngby laboratory will enable the complete core components of the Majorana chip to be manufactured directly on site. This research is based on years of cooperation with leading Danish research institutions,

Next-generation HPE supercomputer offers a mix of Nvidia and AMD silicon

HPE offers three different blades for the GX5000. First is the GX440n Accelerated Blade for mixed-precision computing, with four Nvidia Vera CPUs and eight Nvidia Rubin GPUs. Up to 24 of these blades can be configured per compute rack, which translates to 192 Rubin GPUs per rack. Vera Rubin is

StarlingX 11.0 addresses edge security, IPv4 exhaustion for massive deployments

“We are seeing increasing concern over security at the edge, where physical security is nowhere near as good as in a central office environment,” Waines said. “We are especially seeing this concern from StarlingX users in Europe.” Third-party security testing now assumes physical access to equipment. “It is standard for

Extreme plots enterprise marketplace for AI agents, tools, apps

Extreme Networks this week previewed an AI marketplace where it plans to offer a curated catalog of AI tools, agents and applications. Called Extreme Exchange, it’s designed to give enterprise customers a way to discover, deploy, and create AI agents, microapps, and workflows in minutes rather than developing such components

Serbia Willing to Pay Higher Price for NIS

Serbia is willing to pay a premium to regain control of oil refiner NIS AD as it seeks ways to unshackle the Russian-owned company from crippling US sanctions, President Aleksandar Vucic said on Sunday. NIS’s owners, units of Russia’s Gazprom PJSC, are in talks with third-party investors from Asia and Europe who could potentially take it over, Vucic said in a live broadcast of a government meeting in Belgrade to discuss ways to resolve the dilemma. “If they don’t agree on a purchase price, my proposal is that we offer a better price,” Vucic said. “We are ready to even overpay” for the 56 percent stake held by Gazprom units, he said. Serbia’s only refiner is at risk of running out of crude within days after the sanctions – which took effect Oct. 9 – cut off its supply route through neighboring Croatia. Vucic and Finance Minister Sinisa Mali warned that this could have devastating consequences for Serbia’s economy and its credit rating. The country has just a week to find a solution to avert a fuel crisis, Vucic said. “Whatever it costs, we’ll find the money,” he said. A possible buyout by the state, which holds almost 30 percent of the refiner, would require negotiations with international financiers, and possibly a budget review to secure funding, the president said, without going into a possible valuation. “We want to avoid confiscation, nationalization,” Vucic said. As the majority owner, the Russian stakeholders “have the right to make decisions, but we have the right to live,” he said. Vucic cited Bulgaria’s decision last week to seize control of Lukoil PJSC’s local refinery, also threatened by US sanctions, as an example of what other countries are doing to shield their markets. Trading in NIS shares was suspended in January, after the US Office of Foreign Assets Control unveiled the punitive measures, which

Petronas to File Development Plan for Suriname’s Sloanea Discovery

Suriname’s state-owned oil and gas company Staatsolie Maatschappij Suriname NV (Staatsolie) has declared the Sloanea-1 well in the offshore Block 52 a commercial natural gas discovery and said operator Petroliam Nasional Bhd (Petronas) would now submit a development plan that includes a liquefied natural gas (LNG) facility. “With FID [final investment decision] planned in the second half of 2026, Suriname may expect first gas to flow in 2030”, Staatsolie said. Sloanea-1 is Petronas’ first hydrocarbon discovery in the South American country. It announced the discovery December 11, 2020, after drilling into 4,780 meters using the Maersk Developer rig. The partners had considered Sloanea-1 “commercially unattractive”. They agreed to drill an appraisal well and negotiate a “gas addendum” that would lay out the terms for assessing how to develop the discovery, as announced by Staatsolie’s March 4, 2024. “In accordance with the PSC [the production sharing contract signed April 2013], Staatsolie and Petronas Suriname evaluated monetization options for this gas discovery [Sloanea-1], resulting in a gas addendum to the PSC”, Staatsolie said announcing the declaration of commerciality (DOC). “In 2024, Petronas Suriname advanced the appraisal of the Sloanea reservoir by drilling the Sloanea-2 appraisal well to assess its lateral extent and conduct comprehensive well testing. This confirmed the field’s gas in place and recoverable volumes. “On 11 November 2025, Staatsolie approved the delineation of the commercial field for the Sloanea-1 gas discovery, signifying its ‘declaration of commerciality’. “The selected development concept includes gas development wells, subsea infrastructure and a floating LNG facility, a first in the region”. Malaysia’s state-owned Petronas said separately the DOC marks “Suriname’s first gas development milestone in its expanding deepwater energy landscape”. Petronas has made two more discoveries offshore Suriname, all in Block 52: Roystonea-1 in 2023 and Fusaea-1 and Sloanea-2 in 2024. Block 52 spans over 4,700 square kilometers (1,814.68

Naftogaz Secures Preliminary Deal for Winter LNG from Greece’s DEPA

Greece’s state-owned DEPA Commercial SMSA on Sunday signed a letter of intent (LOI) to supply Ukraine an unspecified volume of liquefied natural gas (LNG) from the United States for the 2025-26 winter via Ukraine’s state-owned Naftogaz Group. This follows Naftogaz’s agreement earlier this month with Atlantic-See LNG Trade SA, formed early November by DEPA and Aktor Group, for the importation of LNG from the U.S. into Ukraine and other European countries. Naftogaz and Atlantic-See “agreed to jointly develop the supply of LNG from the U.S. to Europe and Ukraine through Greek LNG terminals and the Vertical Corridor”, Naftogaz said in a statement on its website November 7. DEPA Commercial chief executive Konstantinos Xifaras said in an online statement Sunday about the LOI, “The supply of U.S. LNG will be facilitated through Atlantic-See, in which DEPA holds a 40 percent stake, underlining the company’s commitment to providing practical and secure energy solutions across Southeast Europe”. DEPA Commercial said, “Under the framework of the prospective agreement, LNG volumes originating from the U.S. are expected to be transported through ‘Route 1’ [of the Vertical Corridor], offered jointly by the gas transmission system operators (TSOs) of Greece (DESFA), Bulgaria (Bulgartransgaz), Romania (Transgaz), Moldova (VestMoldTransgaz) and Ukraine (GTSOU)”. Newly Proposed Gas Routes Recently the TSOs requested their national regulators to approve more flows on the Vertical Corridor, a network of existing gas infrastructure allowing multidirectional flow across seven European countries, via two routes: Routes 2 and 3. “TSOs request the regulators’ approval on the availability of Routes 2 and 3 until April 2026; and the possibility of simultaneous provision of Route 1, Route 2 and Route 3 special capacity products in competing auctions”, said a statement posted on DESFA’s website November 7, announcing a joint letter to regulators. “All participating TSOs have agreed to apply significant

Adnoc Buy of Covestro Wins Conditional EU Approval

Abu Dhabi National Oil Co. won conditional European Union approval for its EUR 12 billion ($13.9 billion) takeover of Covestro AG after it dealt with EU concerns that its state subsidies could stifle competition. The European Commission said Friday that an offer from Adnoc to maintain Covestro’s intellectual property in Europe as well as concessions on the company’s unlimited state guarantee from the UAE settled its earlier fears. Those commitments are valid for 10 years. “Commitments offered by Adnoc effectively address the potential negative effects by allowing market participants to access key Covestro patents in the field of sustainability,” EU competition chief Teresa Ribera said in a statement. “Clear, pre-defined access to these patents will enable others to innovate and advance research in an area that is critical for Europe’s future.” The planned purchase of Covestro would give Adnoc – the biggest oil producer in the United Arab Emirates – control over a German company that supplies materials for some of the world’s most prominent phone and carmakers. Adnoc would own Covestro through its investment unit XRG, set up in last year as the company’s international platform for natural gas, chemicals and energy solutions. In July, the commission, the EU’s antitrust arm, opened a full-scale investigation into the deal under tough new foreign subsidies rules. These are aimed at preventing sovereign states from using their financial muscle to crush competition in the 27-nation bloc. EU officials warned at the time that Adnoc’s state funding may give it an unfair advantage over rivals with less-deep pockets. WHAT DO YOU THINK? Generated by readers, the comments included herein do not reflect the views and opinions of Rigzone. All comments are subject to editorial review. Off-topic, inappropriate or insulting comments will be removed.

The week in 5 numbers: Electricity prices extend rise, regulators rein in data centers

The upper end of Duke Energy’s expanded five-year capital spending plan, which it expects to roll out early next year. Executives attribute the rise in spending to rapid load growth, including many data centers, which they say is likely to continue into the early 2030s. Additional generation added to Duke’s system could exceed 13 GW in the next five years, including 7.5 GW of new gas facilities. Duke is one of many utilities that have bumped their spending in response to projected load growth from artificial intelligence, manufacturing and electrification.

Solar project delays decreased in Q3 2025: EIA

Listen to the article 2 min This audio is auto-generated. Please let us know if you have feedback. Fewer solar developers reported delays in the third quarter of 2025 compared to the same period last year, the Energy Information Administration said in a Monday report. In the third quarter this year, “solar projects representing about 20% of planned capacity reported a delay, a decrease from 25% in the same period in 2024,” EIA said. “Despite the relatively high number of projects reporting delays in 2024, that year was a record year for U.S. solar capacity additions,” EIA said. Developers added around 31 GW of utility-scale solar capacity last year, though their projections at the beginning of the year forecasted 36 GW in additions. Optional Caption Courtesy of Energy Information Administration “Because survey respondents may not anticipate the occurrence or duration of delays, ultimate capacity additions tend to be less than the expected amount that developers report to us at the beginning of the year,” EIA said. The agency said in February that it predicts 32.5 GW of utility-scale solar will be added this year, indicating that less solar may come online this year than last year, despite the decrease in delays. EIA also noted that delays are more common than cancellations, and “less than 1% of planned solar capacity is entirely cancelled in a typical month … Much of the reported delayed capacity occurs at projects that are in the late construction or testing phases just before they come online. These delays are typically only for a month or two.” Justin Baca, vice president of markets and research at the Solar Energy Industries Association, said in an email that it’s “important to note that most of the solar capacity that has come online this year began construction last year.” “The

Arista, Palo Alto bolster AI data center security

“Based on this inspection, the NGFW creates a comprehensive, application-aware security policy. It then instructs the Arista fabric to enforce that policy at wire speed for all subsequent, similar flows,” Kotamraju wrote. “This ‘inspect-once, enforce-many’ model delivers granular zero trust security without the performance bottlenecks of hairpinning all traffic through a firewall or forcing a costly, disruptive network redesign.” The second capability is a dynamic quarantine feature that enables the Palo Alto NGFWs to identify evasive threats using Cloud-Delivered Security Services (CDSS). “These services, such as Advanced WildFire for zero-day malware and Advanced Threat Prevention for unknown exploits, leverage global threat intelligence to detect and block attacks that traditional security misses,” Kotamraju wrote. The Arista fabric can intelligently offload trusted, high-bandwidth “elephant flows” from the firewall after inspection, freeing it to focus on high-risk traffic. When a threat is detected, the NGFW signals Arista CloudVision, which programs the network switches to automatically quarantine the compromised workload at hardware line-rate, according to Kotamraju: “This immediate response halts the lateral spread of a threat without creating a performance bottleneck or requiring manual intervention.” The third feature is unified policy orchestration, where Palo Alto Networks’ management plane centralizes zone-based and microperimeter policies, and CloudVision MSS responds with the offload and enforcement of Arista switches. “This treats the entire geo-distributed network as a single logical switch, allowing workloads to be migrated freely across cloud networks and security domains,” Srikanta and Barbieri wrote. Lastly, the Arista Validated Design (AVD) data models enable network-as-a-code, integrating with CI/CD pipelines. AVDs can also be generated by Arista’s AVA (Autonomous Virtual Assist) AI agents that incorporate best practices, testing, guardrails, and generated configurations. “Our integration directly resolves this conflict by creating a clean architectural separation that decouples the network fabric from security policy. This allows the NetOps team (managing the Arista

AMD outlines ambitious plan for AI-driven data centers

“There are very beefy workloads that you must have that performance for to run the enterprise,” he said. “The Fortune 500 mainstream enterprise customers are now … adopting Epyc faster than anyone. We’ve seen a 3x adoption this year. And what that does is drives back to the on-prem enterprise adoption, so that the hybrid multi-cloud is end-to-end on Epyc.” One of the key focus areas for AMD’s Epyc strategy has been our ecosystem build out. It has almost 180 platforms, from racks to blades to towers to edge devices, and 3,000 solutions in the market on top of those platforms. One of the areas where AMD pushes into the enterprise is what it calls industry or vertical workloads. “These are the workloads that drive the end business. So in semiconductors, that’s telco, it’s the network, and the goal there is to accelerate those workloads and either driving more throughput or drive faster time to market or faster time to results. And we almost double our competition in terms of faster time to results,” said McNamara. And it’s paying off. McNamara noted that over 60% of the Fortune 100 are using AMD, and that’s growing quarterly. “We track that very, very closely,” he said. The other question is are they getting new customer acquisitions, customers with Epyc for the first time? “We’ve doubled that year on year.” AMD didn’t just brag, it laid out a road map for the next two years, and 2026 is going to be a very busy year. That will be the year that new CPUs, both client and server, built on the Zen 6 architecture begin to appear. On the server side, that means the Venice generation of Epyc server processors. Zen 6 processors will be built on 2 nanometer design generated by (you guessed

Building the Regional Edge: DartPoints CEO Scott Willis on High-Density AI Workloads in Non-Tier-One Markets

When DartPoints CEO Scott Willis took the stage on “the Distributed Edge” panel at the 2025 Data Center Frontier Trends Summit, his message resonated across a room full of developers, operators, and hyperscale strategists: the future of AI infrastructure will be built far beyond the nation’s tier-one metros. On the latest episode of the Data Center Frontier Show, Willis expands on that thesis, mapping out how DartPoints has positioned itself for a moment when digital infrastructure inevitably becomes more distributed, and why that moment has now arrived. DartPoints’ strategy centers on what Willis calls the “regional edge”—markets in the Midwest, Southeast, and South Central regions that sit outside traditional cloud hubs but are increasingly essential to the evolving AI economy. These are not tower-edge micro-nodes, nor hyperscale mega-campuses. Instead, they are regional data centers designed to serve enterprises with colocation, cloud, hybrid cloud, multi-tenant cloud, DRaaS, and backup workloads, while increasingly accommodating the AI-driven use cases shaping the next phase of digital infrastructure. As inference expands and latency-sensitive applications proliferate, Willis sees the industry’s momentum bending toward the very markets DartPoints has spent years cultivating. Interconnection as Foundation for Regional AI Growth A key part of the company’s differentiation is its interconnection strategy. Every DartPoints facility is built to operate as a deeply interconnected environment, drawing in all available carriers within a market and stitching sites together through a regional fiber fabric. Willis describes fiber as the “nervous system” of the modern data center, and for DartPoints that means creating an interconnection model robust enough to support a mix of enterprise cloud, multi-site disaster recovery, and emerging AI inference workloads. The company is already hosting latency-sensitive deployments in select facilities—particularly inference AI and specialized healthcare applications—and Willis expects such deployments to expand significantly as regional AI architectures become more widely

Key takeaways from Cisco Partner Summit

Brian Ortbals, senior vice president from World Wide Technology, which is one of Cisco’s biggest and most important partners stated: “Cisco engaged partners early in the process and took our feedback along the way. We believe now is the right time for these changes as it will enable us to capitalize on the changes in the market.” The reality is, the more successful its more-than-half-a-million partners are, the more successful Cisco will be. Platform approach is coming together When Jeetu Patel took the reigns as chief product officer, one of his goals was to make the Cisco portfolio a “force multiple.” Patel has stated repeatedly that, historically, Cisco acted more as a technology holding company with good products in networking, security, collaboration, data center and other areas. In this case, product breadth was not an advantage, as everything must be sold as “best of breed,” which is a tough ask of the salesforce and partner community. Since then, there have been many examples of the coming together of the portfolio to create products that leverage the breadth of the platform. The latest is the Unified Edge appliance, an all-in-one solution that brings together compute, networking, storage and security. Cisco has been aggressive with AI products in the data center, and Cisco Unified Edge compliments that work with a device designed to bring AI to edge locations. This is ideally suited for retail, manufacturing, healthcare, factories and other industries where it’s more cost effecting and performative to run AI where the data lives.

AI networking demand fueled Cisco’s upbeat Q1 financials

Customers are very focused on modernizing their network infrastructure in the enterprise in preparation for inferencing and AI workloads, Robbins said. “These things are always multi-year efforts,” and this is only the beginning, Robbins said. The AI opportunity “As we look at the AI opportunity, we see customer use cases growing across training, inferencing, and connectivity, with secure networking increasingly critical as workloads move from the data center to end users, devices, and agents at the edge,” Robbins said. “Agents are transforming network traffic from predictable bursts to persistent high-intensity loads, with agentic AI queries generating up to 25 times more network traffic than chatbots.” “Instead of pulling data to and from the data center, AI workloads require models and infrastructure to be closer to where data is created and decisions are made, particularly in industries such as retail, healthcare, and manufacturing.” Robbins pointed to last week’s introduction of Cisco Unified Edge, a converged platform that integrates networking, compute and storage to help enterprise customers more efficiently handle data from AI and other workloads at the edge. “Unified Edge enables real-time inferencing for agentic and physical AI workloads, so enterprises can confidently deploy and manage AI at scale,” Robbins said. On the hyperscaler front, “we see a lot of solid pipeline throughout the rest of the year. The use cases, we see it expanding,” Robbins said. “Obviously, we’ve been selling networking infrastructure under the training models. We’ve been selling scale-out. We launched the P200-based router that will begin to address some of the scale-across opportunities.” Cisco has also seen great success with its pluggable optics, Robbins said. “All of the hyperscalers now are officially customers of our pluggable optics, so we feel like that’s a great opportunity. They not only plug into our products, but they can be used with other companies’

When the Cloud Leaves Earth: Google and NVIDIA Test Space Data Centers for the Orbital AI Era

On November 4, 2025, Google unveiled Project Suncatcher, a moonshot research initiative exploring the feasibility of AI data centers in space. The concept envisions constellations of solar-powered satellites in Low Earth Orbit (LEO), each equipped with Tensor Processing Units (TPUs) and interconnected via free-space optical laser links. Google’s stated objective is to launch prototype satellites by early 2027 to test the idea and evaluate scaling paths if the technology proves viable. Rather than a commitment to move production AI workloads off-planet, Suncatcher represents a time-bound research program designed to validate whether solar-powered, laser-linked LEO constellations can augment terrestrial AI factories, particularly for power-intensive, latency-tolerant tasks. The 2025–2027 window effectively serves as a go/no-go phase to assess key technical hurdles including thermal management, radiation resilience, launch economics, and optical-link reliability. If these milestones are met, Suncatcher could signal the emergence of a new cloud tier: one that scales AI with solar energy rather than substations. Inside Google’s Suncatcher Vision Google has released a detailed technical paper titled “Towards a Future Space-Based, Highly Scalable AI Infrastructure Design.” The accompanying Google Research blog describes Project Suncatcher as “a moonshot exploring a new frontier” – an early-stage effort to test whether AI compute clusters in orbit can become a viable complement to terrestrial data centers. The paper outlines several foundational design concepts: Orbit and Power Project Suncatcher targets Low Earth Orbit (LEO), where solar irradiance is significantly higher and can remain continuous in specific orbital paths. Google emphasizes that space-based solar generation will serve as the primary power source for the TPU-equipped satellites. Compute and Interconnect Each satellite would host Tensor Processing Unit (TPU) accelerators, forming a constellation connected through free-space optical inter-satellite links (ISLs). Together, these would function as a disaggregated orbital AI cluster, capable of executing large-scale batch and training workloads. Downlink

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