Note / Introduction: This post is an experiment testing Gemini’s newly released ‘Skills’ feature, specifically the ‘Write Like Me’ profile. To see how well an AI model can mirror my natural writing without relying on prior AI-assisted material, I grounded this skill using my doctoral thesis and a selection of blog posts written in 2022 or earlier. What follows below is the unedited text generated from a raw spoken transcript of my thoughts as I reflected on AI literacy, tool churn, and our ongoing work at UEL.

For those of us working in UK Higher Education, this week brought the quiet release of Gemini’s ‘Skills’ feature into our institutional accounts (arriving right around the 12th of October, if memory serves). As is usually the case with these updates, I have been spending a bit of time working my way around it to figure out what it actually offers beneath the vendor rebranding.

It is a constant challenge for anyone tasked with supporting colleagues and developing institutional AI literacy. The tools are continually marketed as disruptive, but for those of us trying to make sense of them and offer people practical, accessible ways in, the constant shifting of features makes it difficult to pin anything down. I am often reminded of the AI playgrounds I ran back in 2023 for colleagues who had never used ChatGPT. Watching people experiment for the first time and gradually realise the implications for assessment and student work was eye-opening for everyone involved.

My view has always been that the real utility of these tools lies in using them in a grounded, controlled fashion rather than for generic generation. That was why I found custom GPTs so useful: they allowed me to build detailed, focused prompts with embedded reference materials and share them with others so they could benefit from that work without needing to start from scratch. Grounded platforms like Gemini Notebook (formerly NotebookLM) serve a similar purpose, giving staff and students a protected space to interrogate source texts directly with verbatim verification and reframe materials for different educational contexts.

At UEL, we recently went through the process of securing specific permissions to enable Gems within our Google Gemini EDU accounts, including the ability to share them across the institution. The point of doing this was simple: not everyone is a compulsive fiddler like me, nor should they have to be. Most academic and professional services staff do not have the time to invest in working out what every new update means, but they still recognize the importance of getting under the skin of these tools to understand their practical application. Enabling shared Gems meant that those of us willing to experiment could build tailored tools and hand them directly to colleagues to support their day-to-day work.

I was just on the verge of presenting this shareable Gem workflow to our AI literacy working group when the announcement arrived that Gems are being phased out in favour of Skills. It feels like an inevitable cycle in ed-tech, reminiscent of previous shifts across other platforms, but it does make the task of building stable institutional habits feel a bit like aiming at a moving target.

Which brings me back to this current experiment. I decided to test the new ‘Skills’ setup by dictating my thoughts into a speech-to-text transcript and letting the ‘Write Like Me’ skill process the raw audio feed. Grounding the skill in my pre-2022 writing and doctoral thesis ensured that no prior AI-generated text was present in the training sample.

Coincidentally, I received an email yesterday from someone who described my writing and approach as ‘irreverent’. I actually took it as a compliment, though when I looked up the dictionary synonyms, I was amused to find words like ‘disrespectful’, ‘disdainful’, and ‘contemptuous’. I certainly have no desire to be disdainful or contemptuous, but I do not mind being called irreverent if it refers to a healthy skepticism towards administrative tech-churn and overhyped promises. Interestingly, the initial skill profile analysis picked up on that same tone before the email even arrived.

Whether that irreverence or my actual voice comes through in this generated draft is something I will leave to you to decide.

No I won’t: it’s a long way off I think. Just to note GPT Zero assumed it was ‘human written; polished by AI’ . I hope this isn’t what I sound like normally. I worry that it is clearer than I am and undoubtedly it really helps with organising things structurally so for functional things it has really potential utility but the unedited output above is definitely not something I’d be comfortable putting my name to without significant editing. Finally, I had to remove two short self-standing sentences common in (what I assume to be) AI generated posts on Linkedin. It was too uncomfortable. I am, though, going to use this as an attempt to iterate and improve it.

for reference; here’s the raw transcript.

“Okay, let’s try this. This is going to be a short blog post, and it’s basically to test the skills feature. So, for EDU users of Gemini, the skills function in the UK at least was released on the 12th of October, I think it was. Anyway, this week it was released. And I am still working my way around this. As has been the case frequently with so many of these AI tools that like to see themselves as disruptive, but for somebody like me who’s trying to make sense of it and offer people ways in, it’s really difficult to pin anything down because everything keeps changing. I remember back in 2023 running AI playgrounds for people who’d never touched ChatGPT or similar, and just watching as they were experimenting with things, aghast as they were realizing the implications for the ways in which their students were likely to work, and so on and so forth. For me, it’s the obvious areas of utility have always been in the ways in which you can use the tools in a much more grounded and controlled fashion. So, I was an enthusiastic user personally for customized GPTs and very much appreciated the ability to share those custom GPTs with others so that they might use my own efforts at detailed and focused prompts with built-in grounded material. It’s quite clear, probably for cost-based reasons, that the ability to do that was going to change. But I do like the way that you can effectively ground what is now called Google Notebook, formerly NotebookLM. And in EDU spaces, I think it’s really important that people, all stakeholders, that’s both students and staff, are sufficiently familiar with these tools and their utility and function. So, I was delighted when I found out that we were able here at UEL with a specific request and additional permission I would generate. [at this point it stopped transcribing and gave me a fail message. The next bit of recording didn’t work so when I started again, I’d lost my thread a bit]. Okay, I’m trying again, so it failed on the first attempt. Look, what I’m trying to do is get over the point that not everybody is a compulsive fiddler like me. Not everyone is willing to invest time in working out what all the very many changes mean. But a lot of people in positions like me recognize that it’s really important that we get under the skin of LLMs and their functions in order that we might better understand actual utility and application. This is absolutely fundamental. But also the ways in which sophisticated users with complex multi-tool workflows are going to be utilizing these tools to get the most out of them. And I’ve always thought for a long time that customization, whether it was custom GPTs or Claude projects or whatever it is or Copilot agents, were going to be a critical way forward. And agentic AI is something that we have limited ability to control. But I keep changing it and I think that’s largely due to cost reasons. And the reason I’m writing this is because I was all set to tell colleagues in our AI literacy working group that they had of course access to Gems in our Google Gemini EDU accounts, but that we’d enabled for some people sharing functionality. And I think it’s with the sharing that we really start seeing some of the real opportunities. What used to be called NotebookLM, now Gemini Notebook is a good example of how you can ground an LLM, interrogate the text or texts or sources that you include, have verbatim verification opportunities, and can reframe those same documents in any combination that you like into different formats from flashcards to infographics, short animated videos, and what have you. What’s key to me though is that that important distinction that students need to understand and academic colleagues and professional services colleagues need to understand between grounding the LLM in whatever way it happens to be or using a generic tool for free or whatever it is. So more protected spaces like Gemini through the approved system here at UEL, I think is a no-brainer, but it takes a while to encourage people to change their behaviors. And the point I’m trying to make is that I was going to present this thing on Gems and then I suddenly find out that they’re phasing this out in the way that which was inevitable, I suppose. OpenAI custom GPTs, which had such utility. What Google have done is they said, ‘Yeah, we’re getting rid of Gems, but we’re bringing in Skills.’ And literally this week it’s been released, and the first one that caught my eye was Write Like Me. For very obvious reasons, this is something that we need to be worried about, but it also is something that offers potential. So what I did, by way of experiment, is I uploaded my doctoral thesis and a selection of blog posts that I wrote in 2022 or earlier, just so that I could guarantee that there wasn’t any kind of sense of AI assistance creeping its way into any of those documents. And I thought, ‘Let’s give this a go.’ So I don’t know what it’s going to come with, but I am going to post verbatim, and I want this noted in the introduction, the both the final result and the transcript. And let’s have a look at its application of its analysis of my writing style. I got an email yesterday from someone who called me irreverent, and I like that as an adjective. And I thought, I wonder what they meant, and I looked it up, and it was great because the word itself I took as a compliment, but the synonyms that came up were disrespectful, disdainful, scornful, contemptuous, derisive. I don’t want to be any of those things, but I actually don’t think I mind being irreverent. And one of the things that the skill noted, not in so many words, but it noted my tendency towards what might be called irreverence. So let’s see if that comes through.”