2407252121 ideation, planning, analysis and review

an update to making with generative ai

Aug 03, 2024

For all the hype around automated synthetic productivity, generative AI is but one of the technologies in the growing digital landscape we contend with every day. Broadly, there are general-purpose tools and domain-specific applications. In the last twenty-five years, between CAD and generative AI, several technologies have occupied the repertoire aiding architectural imagination. When probed collectively, what is the profession if a tool and skill-based reading of architecture is made? Has automation made knowledge work from the discipline better? In addition to learning to see shifts in spatial discourses through the digital, there is a genuine issue of how sustainable these are.

Like an image editor or word processor, chatbots are general-purpose applications compared to building information modelling (BIM) or computational design tools. Without specific guidelines for domain-specific making, knowing if you use the service efficiently is a conundrum. Even before optimisation is achieved without an aspiration to implement automation, any tool may not yield the desired results. Defining work and the resources to get it done digitally is the task to attend to. Unlike in the past, the profession is more tool-dependent than ever before. Generative AI, thus, is an opportunity to evaluate if all hardware, and software, are in sync and operating at capacity.

If an architect must have WRITING, DRAWING, COORDINATION, and DOCUMENTATION skills, how generative AI addresses productivity concerns for these activities is a multi-stage project to get to. The implementation strategy is to locate the first principles of making for one and then extend them to the rest. Note that the broader writing workflow is detailed. I have arranged it as 📐 ideation, 🗺️ planning, 📊 analysis, and 🔍 review as a mental model to manage the various material text generation instances, writing, irrespective of the requirements of type, is required. The shift here is in the past; we worked with human hallucinations. Now, a machine is doing that for you.

📐 IDEATION. Our time-tested habit of generating concepts is first to read, ideally on the subject or anything around the periphery of the topic we are trying to write about. However, with generative AI, you do not need to follow this process. Ideation is a mixed bag, heavily influenced by an individual's understanding or awareness of the points they want to present. The output of all LLMs is almost similar, highlighting the need for a disciplined approach that machines can augment. This disciplined approach is crucial in using generative AI effectively, keeping you focused and determined to achieve your creative and problem-solving goals.

🗺️ PLANNING. There can be some overlap between ideation and planning, especially in the construction of outlines. Implementing a narrative structure is also staged here. If an agent is introduced, it is in the role of a research assistant. Any service connected to the internet is best suited to audit any resource allocation strategies. Fact-checking and finding additional sources instead of relying on conventional search is recommended. The output of the replies tends to vary when a model is connected to the web against when it is not. It is possible to get multiple perspectives as a result. If a project plan is required, switching back to a static LLM is best.

📊 ANALYSIS. While in the first two stages, it is possible to freewheel and try out outputs from the many models for analysis, the only one built for it is ChatGPT. For any form of data analysis, I recommend only OpenAI's offerings. Google Gemini and Microsoft CoPilot show promise but are expensive to implement and relatively less reliable. Having a conversation on what kind of analysis and even how to conduct them is the productivity boost generative AI has made possible. Generative AI has revolutionised data science, making every stage more efficient and effective. Implementing a custom GPT can further increase efficiency.

🔍 REVIEW. Text output from Claude works well. Its tone is better. This is important when you do not have finetuning skills yet. Llama 3.1 and Mistral have improved. Therefore, alternatives to try. I do consistency checks to fix gaps in a narrative. Reviews are possible at document scale or sections. Both, ideally, can have different workflows. Broad audits require larger LLMs such as Claude. For sections and paragraph scale, it combines Notion AI and Grammarly. The review stage is also when texts are processed for dissemination, e.g., in social media. Summarisation works well everywhere, but finding one that aligns most optimally and sticking with it helps.

For efficiency, the minimum subscriptions are ChatGPT + Llama - ideation, Claude - planning, ChatGPT - analysis, Claude + Notion AI + Grammarly - review. Search services such as Exa and Kagi give access to an alternate internet, which is not Google, Bing or DuckDuckGo. There is no one point of view of the internet today. Your existence is relative to how several algorithms decide to deliver content to you. This is in addition to the subscriptions, free or paid, making up your personal internet. After trying out around 21 AI services, I have narrowed the lot down to 13. The inquiry after this round is to reduce further the services used.

What is an ideal AI-enhanced model for creating work on the internet? This is the question via practice, and I am trying to find an answer. On a personal internet, what tools you end up using are subject to your skills and how much you spend on them. AI is one of the several material types available for knowledge work today. It introduces a layer that either can be ignored or learned how to use for work. To reference previous notes, it is both an information service and an application. Andrej Karpathy calls it software 2.0. An information layer that is also an application is the opportunity presented. How what is built with the technology is appreciated is the returns to expect from the experiment.

References & Links

Data and information documentation is a necessary skill at times. Continuous offline access to personal digital material repositories was essential until 2017, when data and information flows were limited. Websites also used to break back then. With cheaper online storage costs and constantly updated streams to subscribe to new ways of seeing the world is possible. Generative AI is one of the many information services on the internet. Between the previous post and this one, publishing reading lists is a contextualisation of ideas of the moment. The broader discussions these can lead to are left speculative here for now.
The position to read it from is how to introduce generative AI to architecture as a discourse. Before teaching the technology, how it is extendable as an information service and, therefore, as an application needs implementation and discussion. Architectural research extends digital formal investigations, which is the other discussion space. My current trajectory is how disciplinary theory and, eventually, data science inform spatial making. For now, it follows chatter from the tech industry. How tools are used is relative to the task, skill level and status of the subscription. As the space is very volatile, the position is but a benchmark to build from.

Software 2.0

Making Better Decisions with Cassie Kozyrkov, Google's First Chief Decision Scientist

But who in your organization is actually responsible for data quality? Whose job is it to make sure the data is well designed, intelligently collected, curated, managed, and documented?

How Do We Teach AI? In Florida, One Architecture School Offers a Precedent
- When Form Follows Meanings: AI’s Semantic Turn in Architecture
- Narrative, Media, and AI in Architectural Academia; A Conversation with Natasha Sandmeier
  - Win a copy of Joseph Choma's “Études for Architects”, a book of pedagogical design games for architects
- Zaha Hadid Architects: “We were born digital”
- Wait, is PowerPoint cool now? No longer the sole domain of Bob from accounting, the presentation software is being repurposed for parties, pick-up lines and the persuading of parents.
- Global BIM market expected to more than triple to $26.4B over the next decade
- Backed by Microsoft, AWS and Meta, the Overture Maps Foundation launches its first open map datasets
  - Who is Overture for? Overture is for developers who build map services or use geospatial data.
  - The Overture Maps buildings theme describes human-made structures with roofs or interior spaces that are permanently or semi-permanently in one place
- River Travel. An Interview with Anupama Kundoo
- “We get to remind people that at one point, the Olympics actually had the arts as a section that ran all these competitions,” Williams said. “Sculpture, architecture, visual arts. The idea we get to put the arts back in. ... Why not take this moment to bring awareness?”
  - Pharrell advocates for reviving arts competitions for 2028 Olympics at Louis Vuitton event
  - Pharrell Williams Shares His Inspirations, From Henry Taylor to ‘Jabberjaw’ The multifaceted producer, designer and new creative director of men’s wear at Louis Vuitton gives T a glimpse of his creative world.
How Generative AI Can Augment Human Creativity
How Generative AI Is Changing Creative Work
Creativity in the age of generative AI | Nature Human Behaviour
Why I wrote an AI transparency statement for my book, and think other authors should too
- First, has any text been generated using AI?
- Third, has any text been suggested using AI? This might include asking ChatGPT for an outline, or having the next paragraph drafted based on previous text.
- Second, has any text been improved using AI? This might include an AI system like Grammarly offering suggestions to reorder sentences or words to increase a clarity score.
- Fourth, has the text been corrected using AI and – if so – have suggestions for spelling and grammar been accepted or rejected based on human discretion?
Art and the science of generative AI
Enhancing Work Productivity through Generative Artificial Intelligence: A Comprehensive Literature Review
Why agents are the next frontier of generative AI
The Generative AI List of Lists: 5000 Models, Tools, Technologies, Applications, & Prompts
- The 20 best generative AI tools in 2024
Zuck's new Llama is a beast
Today, we are announcing Mistral Large 2, the new generation of our flagship model. Compared to its predecessor, Mistral Large 2 is significantly more capable in code generation, mathematics, and reasoning. It also provides a much stronger multilingual support, and advanced function calling capabilities.
Cloud-native architecture and technologies are an approach to designing, constructing, and operating workloads that are built in the cloud and take full advantage of the cloud computing model.

{spatial_observation}s-drafts

Discussion about this post