I have fully embraced Autumn, this month, with soups, walks on crunchy leaves, and, of course, Halloween celebrations. Oh, and I was interviewed by the BBC, on the use of AI-generated images for scams. Research I joined a group at Sussex, preparing a funding application to create a doctoral centre. And, at the end of the month, I travelled to Barcelona to meet with some great colleagues to work on a research idea for a grant application.  

I was recently interviewed by the BBC about the use of AI-generated images in scam websites. The story followed reports that a small jeweller was being inundated with calls and emails from customers mistaking her legitimate business (Scarlett Jewellery) for a scam website with a very similar name (Scarlett Jewels). One of the features of the scam website is that it uses AI-generated images.

We often assume that emotions play little to no role in business-to-business (B2B) relationships. Rather, decisions in a B2B context are deemed to be purely rational, and to follow a complex process with multiple decision makers. Yet, a metanalysis of research on this topic, conducted by Anne-Madeleine Kranzbühler, Alfred Zerres, Mirella H. P. Kleijnen and Peeter W. J. Verlegh suggests otherwise.

It is widely acknowledged that answers produced by large language models (LLMs) reproduce the biases that informed how they were trained. For instance, assuming that doctors are male while nurses are female. Though, that bias doesn’t simply reflect the systemic inequality in our society. Rather, it reflects the prevalent stereotypes, that is the type of image commonly associated with a particular type of person (namely, male vs female). 

This was a big month on the personal front: I got my British citizenship achieved a key personal goal that I had set for 2025; plus, the kiddo started University. I spent the last 20 or so years wishing for the day when I wouldn’t have to plan work around the school day. I imagined that my productivity would increase dramatically, if I didn’t have to interrupt work because of school runs or after school stuff.

It has been noted by many different people that large language models (LLMs) can, to a certain extent, democratise access to information. Because these models are trained on massive datasets, and use an intuitive prompt-and-response format, they enable users to quickly learn the basics on a given topic or, conversely, do a deep dive on narrow topics of interest. For instance, I have been having an exchange on ChatGPT about “getting into gardening”, which started with general information about things to consider such as light, soil, basic tools, etc… and has now progressed to specific suggestions about what to do at different times of the year, given the area where I live and, even, which local resources and garden centres to visit for inspiration and advice.

This summer was mostly about celebrating several milestones in the kiddo’s life: end of school celebration, leavers’ ball, 18th birthday, A-level results’ day, driving test… But I also enjoyed catching up with friends, reading various fiction and non-fiction books, and swimming lots and lots and lots. Research We finished analysing the first stage of data collection at one of the companies we are working with, for the project on AI agents, and presented the findings to them.

Recently, I was a guest in the podcast “Business Talk”, hosted by Deepak Bhatt. We explored why digital technology is sometimes helpful and sometimes a hindrance, based on the paper “The impact of technology: how features, resources and task demands shape digital well-being”. You can read the blog post about that paper, here. It was an insightful discussion about:

A start-up firm says its small box can help sufferers retain independence and ease relatives' minds.

Authors, websites, and apps can use action intent links to implement "Share on Bluesky" buttons, or similar in-app actions. Logged-in users will be directed to the corresponding action view in the Blu...