https://aibussin.com/tags/prompt-tuning/ Recent content in Prompt Tuning on AIBussin — AI applications, systems and books Hugo en-US Thu, 26 Jun 2025 13:31:41 +0100 https://aibussin.com/post/compiler/ Thu, 26 Jun 2025 13:31:41 +0100 https://aibussin.com/post/compiler/ <p><strong>How to design a pipeline that turns vague goals into smart prompts</strong></p> <h2 id="-summary">🧪 Summary</h2> <p>Why spend hours engineering prompts when AI can optimize its own instructions. This blog post introduces a novel approach toward creating a self-improving AI by treating prompts as programs. Traditional AI systems often rely on static instructions rigid and limited in adaptability. Here, we present a different perspective: viewing the Large Language Model (LLM) as a <strong>prompt compiler</strong> capable of dynamically transforming raw instructions into optimized prompts through iterative cycles of decomposition, evaluation, and intelligent reassembly.</p> https://aibussin.com/post/general/ Thu, 22 May 2025 21:41:54 +0100 https://aibussin.com/post/general/ <h2 id="-summary">🔧 Summary</h2> <p>The <a href="https://arxiv.org/abs/2505.14652">General Reasoner</a> paper shows how we can train LLMs to reason across domains using diverse data and a generative verifier. In this post, I walk through our open-source implementation showing how we built a modular reasoning agent capable of generating multiple hypotheses, evaluating them with an LLM-based judge, and selecting the best answer.</p> <hr> <h2 id="-what-we-built">🧠 What We Built</h2> <p>We built a <code>GeneralReasonerAgent</code> that:</p> <ul> <li>Dynamically generates multiple hypotheses using different <strong>reasoning strategies</strong> (e.g., <code>cot</code>, <code>debate</code>, <code>verify_then_answer</code>, etc.)</li> <li>Evaluates each pair of hypotheses using either a <strong>local LLM judge</strong> or our custom <strong>MR.Q evaluator</strong></li> <li>Classifies the winning hypothesis using <strong>rubric dimensions</strong></li> <li>Logs structured results to a PostgreSQL-backed system</li> </ul> <p>All of this was integrated with our existing stephanie framework, which includes:</p>