https://aibussin.com/tags/cot/ Recent content in CoT on AIBussin — AI applications, systems and books Hugo en-US Tue, 17 Jun 2025 23:31:13 +0100 https://aibussin.com/post/docs/ Tue, 17 Jun 2025 23:31:13 +0100 https://aibussin.com/post/docs/ <h2 id="-summary">📖 Summary</h2> <p>Imagine drowning in a sea of research papers, each holding a fragment of the knowledge you need for your next breakthrough. How does an AI system, striving for self-improvement, navigate this information overload to find precisely what it needs? This is the core challenge our Document Intelligence pipeline addresses, transforming chaotic documents into organized, searchable knowledge.</p> <p>In this post we combine insights from <a href="https://arxiv.org/pdf/2505.21497" target="_blank" class="paper-badge" style="display: inline-block; padding: 6px 10px; background: #f3f4f6; border-left: 4px solid #3b82f6; border-radius: 4px; margin: 4px 0; text-decoration: none; color: #1f2937;"> <strong>Paper2Poster</strong>: Towards Multimodal Poster Automation from Scientific Papers </a> and <a href="https://arxiv.org/abs/2506.10952" target="_blank" class="paper-badge" style="display: inline-block; padding: 6px 10px; background: #f3f4f6; border-left: 4px solid #3b82f6; border-radius: 4px; margin: 4px 0; text-decoration: none; color: #1f2937;"> <strong>Domain2Vec</strong>: Vectorizing Datasets to Find the Optimal Data Mixture without Training </a> to build an AI document profiler that transforms unstructured papers into structured, searchable knowledge graphs.</p> https://aibussin.com/post/arm/ Wed, 28 May 2025 22:22:46 +0100 https://aibussin.com/post/arm/ <h2 id="summary">Summary</h2> <p>Chain-of-thought is powerful, but which chain? Short explanations work for easy tasks, long reflections help on hard ones, and code sometimes beats them both. What if your model could adaptively pick the best strategy, per task, and improve as it learns?</p> <p>The <code>Adaptive Reasoning Model</code> <strong>(ARM)</strong> is a framework for teaching language models how to choose the right reasoning format direct answers, chain-of-thoughts, or code depending on the task. It works by evaluating responses, scoring them based on rarity, conciseness, and difficulty alignment, and then updating model behavior over time.</p>