インサイト INSIGHTS
Articles on generative AI and putting organizational data to work, from the AskDona Engineering / GFLOPS team. Articles on generative AI and putting organizational data to work, from the AskDona Engineering / GFLOPS team.
Insights
From Simple Prompts to Evidence-Based Assessment From Simple Prompts to Evidence-Based Assessment
In production, the first message a user types is rarely a complete task specification. This article looks at how enterprise AI should move from answering isolated requests to constructing the question that must be answered. In production, the first message a user types is rarely a complete task specification. This article looks at how enterprise AI should move from answering isolated requests to constructing the question that must be answered.
From Human Judgment to Globally Accountable AI-Assisted Decision Support From Human Judgment to Globally Accountable AI-Assisted Decision Support
Human judgment is contextual, valuable, and sometimes wrong. This article looks at how AI should support evidence gathering and preliminary assessment in regulated work while accountable people retain authority over the decision. Human judgment is contextual, valuable, and sometimes wrong. This article looks at how AI should support evidence gathering and preliminary assessment in regulated work while accountable people retain authority over the decision.
Japan as One Example of a Worldwide Organizational Challenge Japan as One Example of a Worldwide Organizational Challenge
Critical organizational knowledge becoming person-dependent is not a Japanese problem with a Japanese solution. It is a global problem that takes different forms depending on labor market, industry, and regulatory environment. Critical organizational knowledge becoming person-dependent is not a Japanese problem with a Japanese solution. It is a global problem that takes different forms depending on labor market, industry, and regulatory environment.
RAG Answer Accuracy Depends on What the Model Is Told Not to Answer: Six Principles for System Prompt Design RAG Answer Accuracy Depends on What the Model Is Told Not to Answer: Six Principles for System Prompt Design
RAG answer quality is not determined by retrieval alone. Learn six principles for designing RAG system prompts—how the model uses reference information and when it should not answer—with practical prompt examples. RAG answer quality is not determined by retrieval alone. Learn six principles for designing RAG system prompts—how the model uses reference information and when it should not answer—with practical prompt examples.
Is It Really Tacit Knowledge? Is It Really Tacit Knowledge?
Much of what organizations label tacit knowledge is really recordable fact and definable rule—a Batch Assessment provider's guide to telling them apart. Much of what organizations label tacit knowledge is really recordable fact and definable rule—a Batch Assessment provider's guide to telling them apart.
Improving the Information Behind AI-Assisted Assessments: Lessons from Vulnerability Assessment Improving the Information Behind AI-Assisted Assessments: Lessons from Vulnerability Assessment
When assessment AI returns ‘no information,’ the cause may lie in assessment design, retrieval and registration, evidence and controls, or source and responsibility boundaries. This article explains how to turn those results into targeted improvements. When assessment AI returns ‘no information,’ the cause may lie in assessment design, retrieval and registration, evidence and controls, or source and responsibility boundaries. This article explains how to turn those results into targeted improvements.
Do Not Wait for Perfect Data Before You Start Do Not Wait for Perfect Data Before You Start
Assessment AI does not require an organization-wide data cleanup before the first run. Start with the assessment sheets and evidence you already have, then use incorrect results and missing-information outputs to identify the information that actually deserves improvement. Assessment AI does not require an organization-wide data cleanup before the first run. Start with the assessment sheets and evidence you already have, then use incorrect results and missing-information outputs to identify the information that actually deserves improvement.
Designing for Deployment Beyond the PoC: Patterns We’ve Seen as a Solution Provider Designing for Deployment Beyond the PoC: Patterns We’ve Seen as a Solution Provider
Drawing on GFLOPS’s experience supporting AskDona implementations, this article outlines the conditions that separate projects that stall at the PoC stage from those that move on to scale. Drawing on GFLOPS’s experience supporting AskDona implementations, this article outlines the conditions that separate projects that stall at the PoC stage from those that move on to scale.