Plutonic Rainbows

Today's improvements focused on fixing errors and enhancing the prompt analysis display. I resolved React errors caused by the diff-match-patch library by implementing a local version of the library instead of relying on an external CDN. We then completely redesigned the prompt analysis section to make it more user-friendly, first by simplifying it to a clean, concise summary and then by adding intelligent keyword highlighting that automatically identifies and color-codes important terms like clarity, specificity, and scientific to help users quickly understand what makes their prompts effective. A color-coded legend was added below the summary to explain what each highlight represents. These changes not only fixed the technical issues but also transformed the analysis into a more readable, visually engaging component that gives users immediate insight into their prompt's strengths.

Big temperature drop today. Feels more like March should be. Decided to watch John Carpenter's The Thing, which I haven't watched for a few years. Also bought some albums from Brian Grainger.

Also just discovered Michael J. Blood - Spaces In Between.

Sounds good, especially if you like the work of Actress, Gescom, Larry Heard or Shinichi Atobe.

Made several design improvements to the prompt engineering form, enabling easier data capture and allowing rankings in multiple ways. This will enhance training effectiveness. Additionally, I have included a copy icon to conveniently transfer the text to another agent.

I’ve finally decided to use both Claude Code and Cursor for my editing, as each has strengths in different areas. Claude Code operates via the command line, whereas Cursor is a standalone application, making it beneficial to have both available. For many months, I’ve been waiting for OpenAI to provide complete agent integration, but even now, my preferred application, Nova by Panic, lacks support.

True agent integration likely only became fully available with Cursor about a month ago, and Claude Code itself is also very new, having been released on February 24th.

The rapid advancement of Chinese artificial intelligence has significantly reshaped the global AI landscape, particularly in the Western world. Companies like Baidu, Alibaba, and Tencent have developed sophisticated AI models and applications that have forced Western tech giants to accelerate their own AI development. The emergence of Chinese AI companies offering competitive products at lower costs has created a new dynamic in the market, pushing Western companies to innovate faster and consider more aggressive AI deployment strategies. This influence extends beyond just technology companies, as Chinese AI solutions are increasingly being adopted by Western businesses seeking to maintain competitive advantages in an increasingly automated world.

The impact of Chinese AI development has also led to significant shifts in Western policy and regulatory frameworks. Governments and organizations in the West are now grappling with how to balance the benefits of Chinese AI technology with concerns about data privacy, national security, and technological sovereignty. This has resulted in new regulations and restrictions on Chinese AI investments and partnerships, while also prompting increased investment in domestic AI capabilities. The competitive pressure from Chinese AI development has catalyzed a new era of technological innovation in the West, as companies and governments work to maintain their technological edge while navigating the complex geopolitical landscape of AI development.

I have implemented a fine-tuning system that collects and utilises user feedback to improve prompt refinements. The system stores feedback data (including prompts, ratings, and comments) in a SQLite database, and once I have collected 50 or more high-quality samples (rated 4 or 5 stars), I can initiate a fine-tuning process with OpenAI. This creates a custom model that's specifically tuned to my use cases and feedback.

When users interact with the /refine endpoint, the system now intelligently checks for the availability of a fine-tuned model and uses it if one exists, falling back to the standard GPT-4 model if necessary. I've added new endpoints to manage the fine-tuning process for the processes (/train and /training-status), while maintaining all existing API functionality. However, it's worth noting that the actual fine-tuning occurs on OpenAI's servers and requires OpenAI credits, so the system's effectiveness depends on both the quality of collected feedback and available resources.

I kept encountering persistent errors from the CDN hosting the React assets, which significantly disrupted my workflow. Instead of relying on an unpredictable external service, I took the initiative and dedicated most of yesterday to setting up the assets locally. Although it required substantial effort initially, this proactive step has made my development environment much more reliable and responsive. Not only have I eliminated the dependency on external CDN performance, but I also gained more control over asset management, leading to smoother, faster builds and an improved development experience overall.

I've improved my image generation app to ensure that prompt information is consistently saved with each image. Previously, sometimes the prompt text wasn't being properly written to the image files, which made it difficult to remember what prompts created which images. I've added better error handling around the metadata writing process, with specific fallback options if something goes wrong. Now, even if there's a problem with one method of saving the information, the app will try another approach and clearly log what happened. This makes the whole process more reliable and helps me track exactly what prompts were used to create each image, which is essential for my workflow.

I modified my templates to use a signed URL from cloudfront. Serving signed URLs from CloudFront offers a robust solution for securely delivering private content while enhancing performance. By embedding expiration times, IP restrictions, and custom policies directly into the URL, you can tightly control access and ensure that only authorised users view your content. At the same time, leveraging CloudFront’s global CDN means that content is cached at edge locations, reducing latency and speeding up delivery — protecting your origin resources from direct access and heavy load. This approach not only boosts security but also provides the flexibility and scalability needed for modern, forward-thinking applications.

A day later and I now find that I have settled on Cursor for advanced editing. Cursor is an AI-powered integrated development environment (IDE) developed by Anysphere Inc., designed to enhance developer productivity across Windows, macOS, and Linux platforms. Forked from Visual Studio Code, it incorporates advanced AI features such as intelligent code generation, smart rewrites, and natural language codebase queries. These capabilities enable developers to generate or update entire classes or functions using simple prompts, predict subsequent code edits for efficient navigation, and query the codebase in natural language to retrieve information or refer to specific files and documentation. Cursor also supports the integration of existing extensions, themes, and keybindings, ensuring a seamless transition for users familiar with Visual Studio Code.