Get ONMINE

Stay Ahead, Stay ONMINE

Overcome Failing Document Ingestion & RAG Strategies with Agentic Knowledge Distillation

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

Introduction

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

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

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

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

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

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

Building the pyramid: How Agentic Knowledge Distillation works

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

Overview

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

How to distill documents and build the pyramid: 

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

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

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

Using the pyramid: How it works with RAG & Agents

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

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

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

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

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

Results from the pyramid: Real-world examples

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

Fact-finding (spear fishing): 

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

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

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

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

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

Complex research and analysis: 

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

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

Example response:

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

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

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

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

Example response:

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

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

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

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

Benefits of the pyramid: Why use it?

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

Some of the key benefits we observed include: 

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

Beyond the pyramid: Evaluation challenges & future directions

Challenges

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

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

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

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

Future directions

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

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

Conclusion: Key takeaways and why the pyramid approach matters

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

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

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

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

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

Shape
Shape
Stay Ahead

Explore More Insights

More Articles

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

Shape

Trump extends tariff pause to all USMCA goods

The White House announced Thursday afternoon that it will suspend tariffs on all imports that are compliant with the United States-Mexico-Canada Agreement until April 2. The pause, which was extended to imports from Mexico that adhered to the USMCA earlier Thursday, will now also cover goods from Canada that meet

Read More »

Raizen Is Said to Hire JPMorgan for Argentina Energy Assets Sale

Brazil’s Raizen SA has begun to explore the sale of its oil refinery and network of gas stations in Argentina, according to people familiar with the matter. Raizen, a joint venture between oil supermajor Shell Plc and Brazilian conglomerate Cosan SA, has hired JPMorgan Chase & Co. to manage the sale, said the people, who asked not to be named discussing private matters. Press offices for Raizen and JPMorgan declined to comment.  The energy firm’s potential departure from Argentina would add to a growing list of multinational firms, including Exxon Mobil, HSBC Holdings Plc and Mercedes-Benz, that have chosen to sell operations in the country during the past year despite more investor optimism about President Javier Milei’s economic overhaul.  Brazil’s largest producer of ethanol fuel, Raizen is mulling divestments and slowing down expansions as higher borrowing costs of late in Brazil rattle its finances. Its Dock Sud oil refinery in Buenos Aires is Argentina’s oldest with a capacity of 100,000 barrels a day that only trails two facilities run by state-run oil company YPF SA. Raizen’s network of around 700 gas stations account for 18% of Argentina’s gasoline and diesel sales, second to YPF, which has more than half of the market. The fuel is branded as Shell. Raizen bought the assets for almost $1 billion in 2018 from Shell, which owned them outright, during Argentina’s last experiment with market-oriented reforms. The country then witnessed a period of big government from 2019 to 2023 before voting in libertarian Milei more than a year ago. He is on a crusade to deregulate the economy, in particular the energy and oil sectors. The divestment comes as Milei rips away controls on crude and fuel prices that were used to stem inflation. That was sometimes bad for refiners or drillers, depending on how

Read More »

Data center supply, construction surged in 2024 amid AI boom

Dive Brief: Data center supply in major “primary” markets like Northern Virginia, Atlanta and Chicago surged 34% year-over-year in 2024 to 6,922.6 MW, with a further 6,350 MW under construction at year-end, CBRE said in a Feb. 26 report. The data center vacancy rate in primary markets fell to 1.9%, driving up the average asking rates for a 250-to-500-kilowatt requirement by 2.6% year-over-year to $184.06/kW, reflecting tight supply and robust demand for AI and cloud services, CBRE said in its North America Data Center Trends H2 2024 report. Volume-based discounts for larger tenants “have been significantly reduced or eliminated” due to rising demand for large, contiguous spaces, while data center operators grapple with elevated construction and equipment costs and “persistent shortages in critical materials like generators, chillers and transformers,” CBRE said. Dive Insight: Surging demand from organizations’ use of AI is driving the record data center development, CBRE says. The demand is giving AI-related occupiers increasing influence over data center development decisions like site selection, design and operational requirements. These occupiers are “prioritizing markets with scalable power capacity and advanced connectivity solutions,” the report says.  Demand is also showing up in pricing trends.  Last year was the third consecutive year of pricing increases for 250-to-500-kW slots in primary markets, CBRE said. Following steady single-digit annual declines from 2015 to 2021, average pricing rose 14.5% in 2022, 18.6% in 2023 and 12.6% in 2024. Robust tenant demand, healthy investor appetite for alternative real estate assets and recent interest rate declines are among the factors fueling an exponential increase in data center investment activity, CBRE said. Annual sales volumes reached $6.5 billion in 2024 as average sale prices increased year-over-year, reflecting “the growing scale of data center campuses,” CBRE said. Five transactions exceeded $400 million last year. Notable capital market developments included

Read More »

Bonneville opts to join SPP’s Markets+ day-ahead market over CAISO alternative

Dive Brief: The Bonneville Power Administration plans to join the Southwest Power Pool’s Markets+ real-time and day-ahead market instead of a market being launched by the California Independent System Operator, BPA said in a draft policy released Wednesday. While the CAISO’s Extended Day-Ahead Market may offer greater financial benefits compared to Markets+, overall the SPP market is a better fit for BPA based on market design elements covering governance, resource adequacy, greenhouse gas accounting and congestion revenue, the federal power marketer said. Bonneville expects to make a final decision in May. The BPA’s draft decision sets the stage for the creation of two intertwined day-ahead markets in the West. “The idea that there’s some West-wide market ideal out there that we can get to today is just not something that is on the table,” Rachel Dibble, BPA power services vice president of bulk marketing, said at a press briefing Thursday. “Maybe someday, in future decades, there may be a point where we could merge into one market, but right now, there are many entities who support Markets+.” Dive Insight: The BPA’s decision will have a major effect on market development in the West. It sells wholesale power from federal hydroelectric dams in the Northwest, totaling about 22.4 GW. The federal power marketer also operates about 15,000 circuit miles of high-voltage transmission across the Northwest. The BPA mainly sells its power to cooperative and municipal utilities, and public power districts. In its draft decision, BPA rejected calls to wait for the West-Wide Governance Pathways Initiative to complete its effort to establish an independent governance framework for EDAM.  While a bill — SB 540 — was introduced in the California Legislature last month to implement the Pathways’ second phase, it “limits the availability of full operational administrative independence by requiring that the

Read More »

USA Won’t Hesitate on Russia and Iran Sanctions, Bessent Says

The US will not hesitate to go “all in” on sanctions on Russian energy if it helps lead to a ceasefire in the Ukraine war, Treasury Secretary Scott Bessent said Thursday. Sanctions on Russia “will be used explicitly and aggressively for immediate maximum impact” at President Donald Trump’s guidance, Bessent told an audience at the Economic Club of New York. The Trump administration is pressing Ukraine to come to the table for a ceasefire deal with Russia, and Bessent said additional sanctions on Russia could help give the US more leverage in the negotiations. Trump is ready to finalize an agreement that would give the US rights to help develop some of Ukraine’s natural resources if Ukrainian President Volodymyr Zelenskiy agrees to a tangible path for a truce and talks with Moscow, according to people familiar with the matter. Bessent criticized the Biden administration for not going harder on Russian energy sanctions for fear of driving up gas prices and asked what the point of “substantial US military and financial support over the past three years” was without matching sanctions. The US has paused military aid and some intelligence sharing with Ukraine in an effort to force the US ally to agree to negotiations with Russia over the end of the war. Bessent also said the US would ramp up sanctions on Iran, adding that the US will “shutdown” the country’s oil sector using “pre-determined benchmarks and timelines” and that “Making Iran broke again will mark the beginning of our updated sanctions policy.” The Treasury chief suggested that the US would work with “regional parties” that help Iran move its oil onto the market. One of those countries is likely to be Russia, which signaled earlier this week that it was willing to assist the US in talks with Iran on ending its nuclear

Read More »

Oil Gains on Truce Hopes but Closes Week Lower

Oil’s one-day advance wasn’t enough to rescue prices from a seventh straight weekly decline as the prospect of a temporary truce in Ukraine capped on-again, off-again tariff news that upended global markets. West Texas Intermediate futures climbed by 0.7% Friday to settle above $67 a barrel after Bloomberg reported that Russia is open to a pause to fighting in Ukraine, raising the prospect of a resumption in Moscow’s crude exports. US President Donald Trump earlier pressured the two warring nations to hasten peace talks and the White House signaled that it may relax sanctions on Russian oil if there’s progress. Crude also found support from a weakening dollar and US plans to refill its strategic oil reserve, but still was down 3.9% on the week. The Biden administration’s farewell sanctions on Russia have snarled the nation’s crude trade in recent months, with total oil and natural gas revenue last month falling almost 19% from a year earlier, Bloomberg calculations showed. Russia’s oil-related taxes are a key source of financing its war against Ukraine. A potential reintroduction of Russian barrels to the market comes amid a gloomy period for the supply outlook, as OPEC+ forges ahead with a plan to start reviving idled output in April. Meanwhile, Trump’s trade policies have fanned concerns about reduced global energy demand. “You’re seeing some volatility as people try to interpret what they think is going to happen and what it’s going to mean, but the bottom line is Russia has been able to sell its oil,” said Amy Jaffe, director of New York University’s Energy, Climate Justice and Sustainability Lab. Trump signed orders on Thursday paring back tariffs on Mexico and Canada until April 2. That timing coincides with a date when the president is expected to start detailing plans for so-called reciprocal duties

Read More »

EVOL X Fugro International Women’s Day special

Join Energy Voice News Editor Erikka Askeland who speaks to two high profile energy industry business leaders for International Women’s Day. We speak to Nicola Welsh, UK Country Director at geo-data specialist Fugro alongsideLinda Stewart, Director Marine Geophysical Europe, also at Fugro. Tune in to hear Nicola discuss her route from mining camps in the Australian outback to a senior leadership role while Linda charts her 19-year career journey to become Fugro’s first female director in her role in Scotland. There’s serious discussion about leaning in, the “double bind” and what the IWD 2025 call to “accelerate action” really means. This special podcast also serves and the opening of Energy Voice’s highly anticipated Women in New Energy Event which takes place in Aberdeen in June. Recommended for you Celebrating International Women’s Day with Axis Network’s Emma Behjat

Read More »

Lenovo introduces entry-level, liquid cooled AI edge server

Lenovo has announced the ThinkEdge SE100, an entry-level AI inferencing server, designed to make edge AI affordable for enterprises as well as small and medium-sized businesses. AI systems are not normally associated with being small and compact; they’re big, decked out servers with lots of memory, GPUs, and CPUs. But the server is for inferencing, which is the less compute intensive portion of AI processing, Lenovo stated.  GPUs are considered overkill for inferencing and there are multiple startups making small PC cards with inferencing chip on them instead of the more power-hungry CPU and GPU. This design brings AI to the data rather than the other way around. Instead of sending the data to the cloud or data center to be processed, edge computing uses devices located at the data source, reducing latency and the amount of data being sent up to the cloud for processing, Lenovo stated. 

Read More »

Seven important trends in the server sphere

The pace of change around server technology is advancing considerably, driven by hyperscalers but spilling over into the on-premises world as well. There are numerous overall trends, experts say, including: AI Everything: AI mania is everywhere and without high power hardware to run it, it’s just vapor. But it’s more than just a buzzword, it is a very real and measurable trend. AI servers are notable because they are decked out with high end CPUs, GPU accelerators, and oftentimes a SmartNIC network controller.  All the major players — Nvidia, Supermicro, Google, Asus, Dell, Intel, HPE — as well as smaller vendors are offering purpose-built AI hardware, according to a recent Network World article. AI edge server growth: There is also a trend towards deploying AI edge servers. The Global Edge AI Servers Market size is expected to be worth around $26.6 Billion by 2034, from $2.7 Billion in 2024, according to a Market.US report. Considerable amounts of data are collected on the edge.  Edge servers do the job of culling the useless data and sending only the necessary data back to data centers for processing. The market is rapidly expanding as industries such as manufacturing, automotive, healthcare, and retail increasingly deploy IoT devices and require immediate data processing for decision-making and operational efficiency, according to the report. Liquid cooling gains ground: Liquid cooling is inching its way in from the fringes into the mainstream of data center infrastructure. What was once a difficult add-on is now becoming a standard feature, says Jeffrey Hewitt, vice president and analyst with Gartner. “Server providers are working on developing the internal chassis plumbing for direct-to-chip cooling with the goal of supporting the next generation of AI CPUs and GPUs that will produce high amounts of heat within their servers,” he said.  New data center structures: Not

Read More »

Data center vacancies hit historic lows despite record construction

The growth comes despite considerable headwinds facing data center operators, including higher construction costs, equipment pricing, and persistent shortages in critical materials like generators, chillers and transformers, CRBE stated. There is a considerable pricing disparity between newly built data centers and legacy facilities, reflecting the premium placed on modern, energy-efficient infrastructure. Specifically, liquid/immersion cooling is preferred over air cooling for modern server requirements, CRBE found. On the networking side of things, major telecom companies made substantial investments in fiber in the second half of 2024, reflecting the growing need for more network infrastructure and capacity to accommodate growing demand from AI and data providers. There have also been many notable deals recently: AT&T’s multi-year, $1 billion agreement with Corning to provide next-generation fiber, cable and connectivity solutions; Comcast’s proposed acquisition of Nitel; Verizon’s agreement to acquire Frontier, the largest pure-play fiber internet provider in the U.S.; and T-Mobile’s entry into the fiber internet market via partnerships with fiber-optic providers. In the quarter, Meta announced plans for a 25,000-mile undersea fiber cable that would connect the U.S. East and West coasts with global markets across the Atlantic, Indian and Pacific oceans. The project would mark the first privately owned and operated global fiber cable network. Data Center Outlook

Read More »

AI driving a 165% rise in data center power demand by 2030

Goldman Sachs Research estimates the power usage by the global data center market to be around 55 gigawatts, which breaks down as 54% for cloud computing workloads, 32% for traditional line of business workloads and 14% for AI. By 2027, that number jumps to 84 GW, with AI growing to 27% of the overall market, cloud dropping to 50%, and traditional workloads falling to 23%, Schneider stated. Goldman Sachs Research estimates that there will be around 122 GW of data center capacity online by the end of 2030, and the density of power use in data centers is likely to grow as well, from 162 kilowatts per square foot to 176 KW per square foot in 2027, thanks to AI, Schneider stated.  “Data center supply — specifically the rate at which incremental supply is built — has been constrained over the past 18 months,” Schneider wrote. These constraints have arisen from the inability of utilities to expand transmission capacity because of permitting delays, supply chain bottlenecks, and infrastructure that is both costly and time-intensive to upgrade. The result is that due to power demand from data centers, there will need to be additional utility investment, to the tune of about $720 billion of grid spending through 2030. And then they are subject to the pace of public utilities, which move much slower than hyperscalers. “These transmission projects can take several years to permit, and then several more to build, creating another potential bottleneck for data center growth if the regions are not proactive about this given the lead time,” Schneider wrote.

Read More »

Top data storage certifications to sharpen your skills

Organization: Hitachi Vantara Skills acquired: Knowledge of data center infrastructure management tasks automation using Hitachi Ops Center Automator. Price: $100 Exam duration: 60 minutes How to prepare: Knowledge of all storage-related operations from an end-user perspective, including planning, allocating, and managing storage and architecting storage layouts. Read more about Hitachi Vantara’s training and certification options here. Certifications that bundle cloud, networking and storage skills AWS Certified Solutions Architect – Professional The AWS Certified Solutions Architect – Professional certification from leading cloud provider Amazon Web Services (AWS) helps individuals showcase advanced knowledge and skills in optimizing security, cost, and performance, and automating manual processes. The certification is a means for organizations to identify and develop talent with these skills for implementing cloud initiatives, according to AWS. The ideal candidate has the ability to evaluate cloud application requirements, make architectural recommendations for deployment of applications on AWS, and provide expert guidance on architectural design across multiple applications and projects within a complex organization, AWS says. Certified individuals report increased credibility with technical colleagues and customers as a result of earning this certification, it says. Organization: Amazon Web Services Skills acquired: Helps individuals showcase skills in optimizing security, cost, and performance, and automating manual processes Price: $300 Exam duration: 180 minutes How to prepare: The recommended experience prior to taking the exam is two or more years of experience in using AWS services to design and implement cloud solutions Cisco Certified Internetwork Expert (CCIE) Data Center The Cisco CCIE Data Center certification enables individuals to demonstrate advanced skills to plan, design, deploy, operate, and optimize complex data center networks. They will gain comprehensive expertise in orchestrating data center infrastructure, focusing on seamless integration of networking, compute, and storage components. Other skills gained include building scalable, low-latency, high-performance networks that are optimized to support artificial intelligence (AI)

Read More »

Netskope expands SASE footprint, bolsters AI and automation

Netskope is expanding its global presence by adding multiple regions to its NewEdge carrier-grade infrastructure, which now includes more than 75 locations to ensure processing remains close to end users. The secure access service edge (SASE) provider also enhanced its digital experience monitoring (DEM) capabilities with AI-powered root-cause analysis and automated network diagnostics. “We are announcing continued expansion of our infrastructure and our continued focus on resilience. I’m a believer that nothing gets adopted if end users don’t have a great experience,” says Netskope CEO Sanjay Beri. “We monitor traffic, we have multiple carriers in every one of our more than 75 regions, and when traffic goes from us to that destination, the path is direct.” Netskope added regions including data centers in Calgary, Helsinki, Lisbon, and Prague as well as expanded existing NewEdge regions including data centers in Bogota, Jeddah, Osaka, and New York City. Each data center offers customers a range of SASE capabilities including cloud firewalls, secure web gateway (SWG), inline cloud access security broker (CASB), zero trust network access (ZTNA), SD-WAN, secure service edge (SSE), and threat protection. The additional locations enable Netskope to provide coverage for more than 220 countries and territories with 200 NewEdge Localization Zones, which deliver a local direct-to-net digital experience for users, the company says.

Read More »

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

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

Read More »

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

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

Read More »

2025 playbook for enterprise AI success, from agents to evals

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

Read More »

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

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

Read More »
More Articles