I uploaded a collection of my own books and a few articles to Google’s NotebookLM, which deploys RAG (Retrieval-Augmented Generation) techniques to interrogate any corpus. I entered the books in chronological order, as I was interested in how my ideas and themes might have changed over the period of their writing (3 decades!). I prompted…More
LLMs and the web
Early web browsers (e.g. Mosaic in the 1990s and its successors) relied on centralised servers that pre-processed (“crawled”) web pages to identify key terms, words and groups of word. The servers would then add these terms to an index optimised for quick lookup, linking the terms to the URLs of the pages in which these…More
The Quid Pro Coaster
Designers, writers, and illustrators in general are skilled at drawing inspiration from just about any source. As I have shown in previous posts, LLMs seem capable of something similar, especially with source texts that are colourful, include spatial cues, characters, situations and accounts of interesting experiences. But how do they handle dull, prosaic, non-spatial texts,…More
Dialectic between the verbal and the visual
In the 1980s, the architect, theorist and teacher Bernard Tschumi penned an essay “Spaces and Events,” which later appeared as a chapter in his influential book Architecture and Disjunction. In it he outlines an approach to architectural design, at least in a studio teaching context, that elides the literary with the pictorial, to the extent…More
From text to image via LLM
Text and writing are important components in creating architecture. To put it more strongly: text is deeply intertwined with the production of architecture, serving as more than a mere communication tool. Text impacts design thinking, theory, history, and the way the built environment is constructed — materially. As in my previous posts, I’m seeding this…More
AI briefs an architect
Inspired by the game design scenario of my last post (Share your expertise) I asked NotebookLM how the ideas outlined in the collection of sources (e.g. my own published writing) could inform the creation of a physical architectural work. Here’s what it said: Informing Architectural Design The sources provide a wealth of concepts and frameworks…More
Share your expertise
Google’s NotebookLM utilizes the RAG methodology for document interrogation, allowing users to gain insights from multiple texts simultaneously, while the University of Edinburgh has introduced its ELM built on ChatGPT for educational use. Both tools enhance understanding in specialized fields by enabling easy document uploads and tailored responses, benefiting users in architecture, law, and healthcare.More
Smashing the context window
In natural language processing LLMs, the “context window” is the range of text around a target word or phrase that constrains the sequence of text processed in calculating positional and attentional encoding, and hence prediction of the text that follows. A larger context window allows the AI model to capture more subtle relationships and dependencies…More
AI profiles you
In 2009, an academic received feedback through a 360-degree process, similar to analyzing their blog content with AI. The blog, started in 2010, includes 741 posts exploring technology, culture, and design. Using AI, the author inferred insights on personality and product interests, reflecting their intellectual pursuits and the evolution of digital analysis.More
Post-creative AI
There’s a phase of production after an author completes a manuscript (draft or final) when the author might need to compile an index, summaries for publicity, and other “extra-authorial tasks”? The main creative task of writing is complete and the tasks that follow could be undertaken by editors or others, or even an AI. In…More