Biography of an obscure tech exec

A tale and a tool for building AI literacy

Last week was a busy week as I shuttled across the Tasman to speak at back to back conferences in Melbourne and Auckland. For my international readers (57% of DRDers are outside AU/NZ!), no there is no bridge between Australia and New Zealand. It’s roughly a four hour flight and yes you need a passport. Although it has never ceased to make me chuckle that Melbourne Airport at one point just after the pandemic had three queues for passengers: domestic, international and trans-Tasman.

One of the endless delights of my time at Xero was how many useful and fascinating applications of data and AI there are in small business accounting. So many that although I moved on from Xero a while ago, I’m still very much active in the space. Lots of accounting and bookkeeping firms are adopters of AI and still more are keen to learn. So it was my pleasure to speak at the inaugural AI in Accounting Summit hosted by Aider & Karbon.

The brief was to make Gen AI engaging and understandable. I also wanted to ground the past two hurly burly years in the broader seven decade AI story and leave folk with some mental models they can use to hype check new tools and conversations. I had a few cracks at that, this one definitely landed.

As you will see, it’s industry agnostic and instantly reusable. If it helps you, please borrow and repurpose as you spread pragmatism and increase understanding!

A three paragraph biography from Claude

Who hasn’t Googled themselves? After ChatGPT came on the scene, asking your LLM of choice for a bio was a pretty natural extension. This is something I’ve done every quarter or so since Nov 2022. No, I’m not that insecure, it’s an eye opening and grounding experience. Bear with me.

In the below I’ve used the free version of Claude (3.5 Sonnet) because that was open on a browser window when I was writing the talk. Other LLMs will be comparable, use your preferred flavour.

Before you scroll any further, take a second read of the Claude-generated bio. Some of you will know me well, others will only know me through this Substack. Form an overall impression of the bio and then for bonus points, take a punt at picking something that definitely right. And something that’s definitely wrong. I’ll wait while you Google me.

I’m not a celebrity but I’ve got a digital presence

I’ll take a moment at this juncture to point out that I’m not a celebrity. However I’ve got a reasonably extensive online presence for a normal human. I’ve got a full LinkedIn profile. I sit on boards and mentor startups so multiple bios exist online. I’ve got a publication record from my years in academia. I’ve been speaking at recorded public events for over a decade. I write this Substack. I’m not shy of the microphone so I’ve recorded (quickly counting on my fingers) around 10 podcasts over the years, stretching back to circa 2017. If I didn’t want to write my own bio for whatever reason, I could realistically expect to pay someone to do it for me and, based only on what they could find on the public web, expect some useful copy.

So how did Claude do?

When I did my first quick scan of the bio, it seemed pretty plausible. I imagine you thought so too, even if you know me pretty well. Then I went back and read it again and the picture wasn’t as rosy. In the image below, green is correct, orange is sort of a half pass - plausible but not really on the money, red is wrong.

The really stunning one is the second paragraph which is more wrong than right. But oh, isn’t it plausible?! And all so grammatically correct.

For those who are curious, I’ll quickly run through the orange and red.

‘held senior leadership positions in data and analytics’. A lazy journalist might write that, a more awake one, particularly writing in 2025 would say ‘in data and AI’.

Yes I was an Executive General Manager (a fairly unusual title so kudos to Claude) but the ‘Data Science’ bit is pure fabrication.

‘initiatives to help small business …’ yeah plausible but so beige! And not language I have ever used or seen used to describe my contribution while at Xero.

The working at ANZ Bank bit is pure fabrication. And for those outside Australia, it’s particularly odd because ANZ Bank … doesn’t generally call itself ANZ Bank. Kinda like Cher and Madonna, it only needs one name: ANZ. I actually suspect that is part of the problem for the poor LLM. Even humans sometimes get tripped up in text or speech by the common double use of ANZ - for both the Australia and New Zealand region, often a sales territory, and the Australian and New Zealand Banking Corporation, the fourth largest Australian bank at a healthy $86 billion market cap.

The last paragraph is weird. I do indeed believe in the importance of diversity in tech leadership but when I paused and thought about it, I can only remember speaking at one industry conference on that specific topic. So that’s a fail.

And I have three degrees (which yes, could be seen as a tad excessive) but they are all in physics.

In Claude’s defence

A little cheekily, I haven’t yet given you Claude’s complete response.

Yup, it’s the increasingly standard get out of jail free card disclaimer. Don’t get me wrong, I’m pleased to see it. Claude and other LLMs are at pains to point this out, at least in their public chatbots, and I believe they are right to do so. But. It’s a bit like those privacy notices / T&Cs / cookie selection popups. You just go blind to it after a while.

So what?

There was a really interesting reaction to this example at the conference, and it was just the one I was hoping for. One specific attendee summed it up well

“But I just looked up your LinkedIn profile and it’s so wrong!”

Incredulous conference participant who thought Claude did a shocking job

Well yes. Precisely.

The large language model is not searching. It is generating.

Generation can work astoundingly well where there is a great density of data. I bet Halle Berry gets a much more accurate three paragraph bio when she tries this exercise. But there is nothing like generating a biography for a relatively obscure person you know well to see statistical dice rolling in action.

But far too many people unconsciously think that tools like Gemini, ChatGPT, Claude and DeepSeek are basically doing a clever and complex lookup. ‘Web grounding’, retrieval augmented generation and hybrid offerings like Perplexity muddy the waters on this. Hybrid solutions are a fine idea and I’m sure we will see more. But having clarity on the fact that it’s a hybrid is important. LLMs are a great tool but an incomplete answer to almost everything, whether you patch over the gaps with a human or with another complex system. I don’t expect that to change.

Now what?

Knowledge is power and understanding leads to better decisions, personal and professional.

If you found this exercise any combination of enlightening and/or engaging, I hope you will reuse it in your own efforts to increase the Gen AI literacy levels of those around you. Our need for savvy and discerning users is growing by the day.

At a minimum, get Claude to write a three paragraph bio of you right now and always have a great party trick in your back pocket.