I recently completed with the help of Claude – an interactive visualization of moving dots that form different geometric patterns. What makes this particularly satisfying is that I managed to create it mostly through vibe coding.

The Inspiration

I first came across this mesmerizing pattern a few years ago and could not anymore find the original source. Reverse google search though revealed it was also on Reddit.

I wanted to recreate ifor my class. I wanted to make my students guess the underlying patterns as these dots moved. When I teach, I teach mostly through frameworks – always insisting to my students that different people can interpret the same events differently because of their varying perspectives. Depending on which patterns you focus on, you see completely different shapes emerging from the same moving dots.

Vibe Coding with Claude

Starting with just a basic concept, I asked Claude to help me build a heptagram. I asked it then to find 4 triangles with the same dimensions with the vertices along the perimeter of the heptagram. For the longest time, it seemed impossible. I kept tweaking my prompts, explaining what I wanted, but couldn’t quite get there.

I then changed my approach, prompting about dots moving along its perimeter. The initial implementation was straightforward. To help me make sense of it, I asked Claude to implement controls for:

• Adjusting the number of dots
• Changing their speed
• Allowing users to connect dots to form shapes

The real challenge came when I wanted to add triangles and squares that would be overlaid as the dots moved. The first breakthrough came when I realized we needed to treat the heptagram as if it were a flat line with dots spaced proportionally along it. The triangles emerged but they did not look good. At first, the triangles formed by the dots would constantly change size, which wasn’t what I wanted.

After much experimentation, the second breakthrough came when I realized that the key was having the dots slow down near the vertices of the heptagra.

The Result

What you see now is a fully interactive visualization that my students can explore. They can experiment with different settings, observe how the patterns form and transform, and develop their intuition about multiple perspectives. It beautifully illustrates how the same underlying data can yield different interpretations depending on what you choose to focus on.

Despite having no coding background, I was able to recreate a complex mathematical visualization that I can now use in my classroom.

Vibe coding is really becoming a thing.

There's a new kind of coding I call "vibe coding", where you fully give in to the vibes, embrace exponentials, and forget that the code even exists. It's possible because the LLMs (e.g. Cursor Composer w Sonnet) are getting too good. Also I just talk to Composer with SuperWhisper…

— Andrej Karpathy (@karpathy) February 2, 2025

I am currently writing a book chapter on the data economy for training AI models in healthcare. I was curious whether these research paper writing tools would be of any help or I can just rely on the foundational models like Claude, ChatGPT or Gemini for generating ideas and rephrasing words.

I searched google for these different paper writers and came across the following. I then asked them to answer: What is the regulatory landscape in medical data sharing?

Here is my evaluation of these tools, ranked based on their usefulness.

Undermind

I like the follow up questions it asks to make the paper search much more targeted. I also like that they show the timeline of papers and categorize them into themes. This seems to be the best to gain an overview of a field and creating an outline.

Scite

It writes the text already formatted in paragraphs. The citations are displayed on the right, making it easy to verify them and access for further reading. As for the content itself, it more or less makes sense but it seems arbitrary which topics it covers, what papers it cites and how the ideas flow.

Consensus

This tool gives a good overview of different issues and is formatted more like an outline.

Elicit

It provides brief summaries. My guess is that this is more for binary questions like will lead lower my IQ?.

Jenni

It takes an interesting approach by autocompleting the next sentences. I think this is best for writing when you already know what you want to say and just looking for aid with finding relevant papers.

Paperpal

The text generator still needs some improvements. In the example, it mentioned Africa out of nowhere for instance.

Hyperwrite

Not really for scientific papers

Genei

Not sure what’s supposed to happen here. There was just ways to upload your papers, but this can already be easily done with Claude, NotebookLM and ChatGPT

I’m contributing to a class on data-driven digital transformation in the coming semester. For this then, I have to be updated with the latest research on data.

Data governance and digital innovation: A translational account of practitioner issues for IS research – discusses the main challenges of data governance and potential research themes for the IS community:

• Ensuring that data governance protects organizations without hindering innovation
• Finding the repertoires of mechanisms for data governance
• Apart from designing them, understanding the enactment of data governance
• Understanding how organizations can implement digital services even when the data assets are in flux

The role of data for AI startup growth – With the proliferation of generative AI like ChatGPT and Midjourney which have been built generally with public data, the primary way that new startups can differentiate themselves is by training their algorithms with proprietary data. This study verifies that, for AI startups, having proprietary training data is positively correlated with obtaining future VC funding.

Interoperability in the era of digital innovation: An information systems research agenda – Being involved in Open Science and FAIR data, one of the top issues is interoperability which refers to the ability to exchange and understand information from another system. The review provides an excellent overview of the challenges of interoperability and how to increase it (what unit of analysis, which actor is involved and whether it is at the level of IT, data or software). To guide researchers, the review also recommends areas to further study including its conceptualization, scoping and methodology.

The Digital Revolution, 3D Printing, and Innovation as Data – previously, firms “innovated from data” – acquiring and analyzing various consumer data to guide their innovation activities. However, “innovation as data” is reversing this where consumers use digital tools to transform data into innovative products.

Organizations Decentered: Data Objects, Technology and Knowledge – quite a difficult paper given my lack of knowledge in such tradition, It argues that data objects are becoming more prominent, increasingly influencing how organizations work especially in their processes of knowing. This then leads to the decentering of organizations where the relation of domain knowledge over data production becomes looser and internal versus external references are reordered.

Legitimating Illegitimate Practices: How Data Analysts Compromised Their Standards to Promote Quantification – fascinating study of how data analysts made previously illegitimate practices such as using hacky code, not holding out samples and using inaccurate data more acceptable. They did this by standardizing these practices, instantiating the practices in their technology platforms and educating collaborators in various meetings.

I have not blogged in a long time mostly because of my disillusion with AI. I felt like AI (with tools like ChatGPT and MidJourney) can do many of the things I do in my blog much better than I can. This made me forget that the main reason I made this blog is not really to create the most interesting content but mostly for myself – to help me learn things. So now, I’m planning to get back to the rhythm and post more frequently once again.

Soft but Strong: Software-based Innovation and Product Differentiation in the IT Hardware Industry – I like listening to the Decoder podcast where they interview CEOs from many consumer product companies. It is interesting to hear that many hardware companies have more engineers working on software than hardware. This study then validates why this might be the case. Researchers find that “investing in software-based innovation not only develop more new products but are also more likely to launch new products that are distinct from those of rivals as well as their own.”

The role of prototype fidelity in technology crowdfunding – Explores how much prototype fidelity, referring to how the prototype reflects the final product’s look and functionality, affects crowdfunding performance. The study finds that too much fidelity may not be ideal as it restricts the ability to co-create with the community. Moreover, the researchers found that materiality also plays a role in how such prototypes are judged (e.g. purely mobile apps vs. smart devices). There is an assumption that purely digital artifacts would be always updated and so ensuring fidelity of the prototype may not matter too much.

From theories to tools: Calling for research on technological innovation informed by design science – External observers of innovation studies would assume that we mostly spend our time evaluating various tools from design science such as the Business Model Canvas or coming up with new tools for design. But, this is not really the case, in fact, I would only know about these approaches recently as I began teaching more about innovation. This editorial from Technovation calls for much research in the area.

Transitioning additive manufacturing from rapid prototyping to high-volume production: A case study of complex final products – Explores how 3D printing can be scaled in an organization. While it can be easily assumed that technology characteristics such as slow printing speeds or limitations in capacity would be the main barrier, the study provides an alternate view. They find that misaligned technology development processes across subsystem units in an organization can hinder it more. To facilitate scale-up then, synchronization between these disparate teams is the priority.

Reconceptualising innovation failure – advances three dimensions of failure:

• Failure as experimentation – on-going testing towards new iterations of an idea
• Failure as judgment – often a strategy used by managers to reset towards a new direction
• Failure as event – refers to unexpected shocks that require recovery from the organization

From Bits to Atoms: Open Source Hardware at CERN – a new paper from my colleagues at ESADE theorizing about how open source can happen in hardware. Compared to open-sourcing in software, an important concept that they introduce is that of malleability – how much one can modify the artifact, taking into account embodiment (component’s physical or non-physical state), modularity (relationship across these components) and granularity (ability of an object to be decomposed into components). Based on these dimensions then, they speculate that open-sourcing or traditional hardware development may be more likely.