Plutonic Rainbows

I have updated the app to now calculate a weighted score (using a Bayesian average–style formula) for each prompt and display the results on the admin dashboard. This version of the app continues to provide personalised prompt generation, user authentication, rating submission (on a 1-to-10 scale with an optional comment), and an admin dashboard that shows aggregated data (with the top three results highlighted in the table).

Today, I integrated several new features into my prompt evaluation app to enhance its functionality and data-driven personalisation. I expanded the rating scale from a simple binary choice to a 1-to-10 scale, allowing for more nuanced feedback from users. I also updated the prompt generation process to keep prompt text and seed words separate, ensuring that the descriptive content is richer and more varied. Additionally, I incorporated a trained machine learning model into the prompt selection function — this model weights candidate one-liner prompts based on historical user ratings, and it gracefully falls back to random selection if the model isn’t available.

I further enhanced the admin dashboard to provide a clear visual representation of prompt performance. Using Chart.js, I set up a bar chart that displays the average rating and count for each prompt description, and I implemented a feature to highlight the top three performing prompts with distinct colours. These updates, along with maintaining user authentication and basic administrative routes, have made my application more sophisticated and have paved the way for future improvements in personalisation and data analysis.

Later in the day, I added a comments section for users — this will also be saved to the SQL database.

Today, I integrated several key features into my prompt evaluation app. I enhanced the application by implementing dynamic prompt generation with AJAX, transitioned from CSV storage to an SQLite database using SQLAlchemy, and added an admin dashboard complete with Chart.js visualizations to review evaluation data. I also introduced user authentication with Flask-Login, enabling secure registration, login, and logout functionality, and I improved navigation by adding clear links between the prompt evaluation page and the admin dashboard.

Persico's work on the album opens with a burst of real-world textures that blend found sounds, birdsong and ambient recordings into a soundscape that feels both contemporary and steeped in history. On tracks like The Center Cannot Hold and Brutal Threshold, her inventive layering of pebbly noises and breathy vocal drones creates an impression of communing with the past, all while utilising modern sound manipulation techniques.

Throughout the album, Persico, with help from Belgian sound designer Koenraad Ecker, transforms everyday recordings into rich tapestries of audio that hint at sacred, folk and operatic traditions. This process reveals the hidden energy of old spaces, as elements such as distant police sirens and corroded metallic scrapes gradually expose the deep, archival layers embedded in the recordings.

The album continues this experimental journey by exploring various sonic territories on tracks like Maze, Rashid Karami, Kairos and Voices Organ. Whether it's the evocative call to prayer merging with environmental sound or the industrial rhythms and meditative mantras that punctuate the pieces, each track offers a fresh perspective on how the legacy of historic spaces can shape and inspire the modern auditory experience.

You can purchase the album here.

Thankfully, just a few days after it was announced for professional users in the United States, the European Union has finally gained access to this new technology. I tried it out today and found it exceptionally impressive. It actually employs the full 03 model, which has yet to be released for public use. The power of this new technology is that it can search the entire web and collect real-time information before reasoning about that information.

I have built a lightweight Flask web application featuring a binary evaluation system with 'Yes' and 'No' buttons and integrated AJAX to dynamically load new prompts without reloading the page, while storing user feedback in a CSV file. I enhanced the interface by splitting the prompt into two distinct boxes — one for the description and one for the seed words — styling the seed words in bold without a redundant label, matching the font sizes, and adding complementary border colours, all while retaining a hidden comment box for future feedback. Moving forward, I plan to expand the list of available prompts and transition to using SQL for storing the results.

OpenAI has unveiled Deep research, a new AI agent integrated into ChatGPT, designed to autonomously conduct multi-step internet research for complex tasks. Leveraging the upcoming OpenAI o3 model, this tool is optimized for web browsing and data analysis, enabling it to find, analyze, and synthesize information from various online sources, including text, images, and PDFs. This capability allows deep research to generate comprehensive reports in approximately 10 minutes, a task that would typically take a human researcher several hours.

Deep research is particularly beneficial for professionals in fields such as finance, science, policy, and engineering, who require thorough and precise information. Additionally, it caters to discerning consumers seeking personalized recommendations for significant purchases like cars, appliances, and furniture. Each output is fully documented, providing clear citations and a summary of its reasoning process, which facilitates easy reference and verification of information. This feature is especially effective in uncovering niche, non-intuitive information that would typically require browsing numerous websites.

While deep research marks a significant advancement in AI capabilities, experts advise caution in relying solely on AI-generated outputs without human verification. The tool is currently available to ChatGPT Pro subscribers in the U.S., with plans for broader access in the future. OpenAI acknowledges potential limitations, such as the AI’s ability to distinguish authoritative information from rumors and accurately convey uncertainty. Despite these challenges, deep research represents a substantial step toward OpenAI’s broader goal of developing artificial general intelligence capable of producing novel scientific research.

Darkness Darkness's debut album, Animation, is a richly satisfying exploration of extended deep cuts and machine-soul fragments, living up to every bit of its early promise. Issued by the mysterious imprint A Visiting Link as their first full-length vinyl release, it arrives in the form of an eight-track LP that showcases a range of oddball downtempo electronics — an ideal fit for anyone drawn to Actress, Andy Stott, Purelink, False Aralia, West Mineral, or 3XL.

Comparable in spirit to classic works by The Connection Machine or Actress, the album's immersive flow runs from start to finish. Two ten-minute-plus compositions anchor the set: the stuttering, cavernous techno dub of the title cut and the febrile dream sequence of Beauty of the World. They're complemented by shorter, more abstract pieces such as the choral pad play on Exp Sample, the lustrous harmonics of Smile, and the glowingly somnambulant techno-tronics of Empresszzzz. Written and produced by Darkness Darkness at Magnetic Dress and Dedicated Computer Realm, Animation forges a hypnotic path through spacious sound design and reflects an artist with a keen ear for detail — highly recommended for those in search of deeper musical frontiers.

The new high variant of the o3-mini from OpenAI took only 33 seconds to model this simulation, featuring a hundred colourful balls bouncing inside a rotating sphere. Each ball stores a brief record of its recent positions and draws line segments between them with gradually increasing opacity, creating a graceful fading trail. Collision detection is handled by checking whether a ball’s centre (plus its radius) has moved beyond the container sphere’s boundary, after which its velocity is reflected accordingly. This is seriously impressive stuff.