My LLM Exploration Journey and Current Workflow
摘要
我的 LLM 使用從 Copilot Pro 歷經 Gemini、Perplexity、Claude，現主力為 Gemini Advanced（2.5 Pro）。互動模式從問答演進至深度搜尋。現透過 Gemini Advanced 深度搜尋與整理報告，需視覺化時則用 Claude AI（3.7 Sonnet）產出 HTML，滿足探究需求。
Abstract
My LLM journey spanned Copilot Pro, various Gemini versions, Perplexity, and Claude, now centering on Gemini Advanced (2.5 Pro). Interaction evolved from Q&A to deep search. Currently, I use Gemini Advanced for deep searches and initial reports, turning to Claude AI (3.7 Sonnet) for visualized HTML outputs to satisfy in-depth inquiries.
我的大型語言模型探索之旅與當前工作流程
演進的足跡 The Evolutionary Path
我的大型語言模型（LLM）使用歷程，始於 Copilot Pro（基於 ChatGPT 3）。隨後，為了尋求更進階的功能與回應品質，我輾轉試用了數個不同的平台，包括搭載 Gemini 1.5 Pro 的 Gemini Advanced、接著是搭載 Gemini 2.0 Pro 的版本、整合了 Claude 3.7 Sonnet 與 Gemini 2.0 的 Perplexity AI，以及獨立運作的 Claude AI（搭載 Claude 3.7 Sonnet）。如今，我的主力工具已轉變為搭載最新 Gemini 2.5 Pro 的 Gemini Advanced。
My journey with Large Language Models (LLMs) began with Copilot Pro (based on ChatGPT 3). Seeking more advanced capabilities and response quality, I subsequently transitioned through several different platforms. This included Gemini Advanced running Gemini 1.5 Pro, followed by a version running Gemini 2.0 Pro, Perplexity AI which integrated Claude 3.7 Sonnet and Gemini 2.0, and the standalone Claude AI (featuring Claude 3.7 Sonnet). Currently, my primary tool has evolved to Gemini Advanced, powered by the latest Gemini 2.5 Pro.
互動模式的轉變 Transformation in Interaction
與此同時，我與這些 AI 工具的互動模式也經歷了顯著的演變。從最初主要進行單純的一問一答，我逐漸進入了更注重「推理」（Reasoning）能力的階段（Perplexity AI 與 Claude AI）。如今，我更側重於利用「深度搜尋」（Deep Search）功能來進行廣泛且深入的資訊挖掘，而 Perplexity AI 與 Gemini Advanced 在此階段扮演了關鍵角色，協助我更全面地探索問題。
Concurrently, my interaction style with these AI tools has also undergone a significant evolution. Starting primarily with simple question-and-answer exchanges, I gradually moved into a phase emphasizing 'Reasoning' capabilities (Perplexity AI and Claude AI). Now, my focus has shifted towards leveraging 'Deep Search' functionalities for extensive and in-depth information gathering. Both Perplexity AI and Gemini Advanced have played crucial roles in this stage, helping me explore topics more comprehensively.
當前的最佳實踐 Current Best Practice
經過這番探索，我最終確立了一套能夠極佳滿足由好奇心驅動、進行深入問題探究的解決方案。我的標準作業流程是：
- 首先，利用 Gemini Advanced（搭載 Gemini 2.5 Pro）針對特定問題進行數次的「深度搜尋」，以廣泛收集多元觀點與資訊。
- 接著，我會再次運用 Gemini Advanced 將這些分散的搜尋結果進行彙整、提煉，並生成一份統一的基礎 HTML 格式報告。
- 然而，如果報告中需要較為複雜或特定的視覺化呈現時，我會轉而使用 Claude AI（搭載 Claude 3.7 Sonnet）來生成最終的 HTML 報告，以確保達到更理想的視覺化效果與呈現方式。
Through this exploration, I have finally established a solution that effectively satisfies my curiosity-driven need for in-depth inquiry. My standard workflow is as follows:
- First, I utilize Gemini Advanced (powered by Gemini 2.5 Pro) to conduct multiple 'Deep Searches' on a specific question, aiming to gather a wide range of perspectives and information.
- Subsequently, I employ Gemini Advanced again to consolidate and refine these scattered findings into a unified, basic HTML report.
- However, should the report necessitate more complex or specific visual elements, I then turn to Claude AI (running Claude 3.7 Sonnet) to generate the final HTML output, ensuring optimal visualization and presentation.