WTF is GPT and why should I care?

Yesterday, I had the pleasure of talking to a leading UK retail business about all things GPT. I have a reasonably close relationship with the company, and was asked — as a favour — to shed some light on ChatGPT’s usefulness in the context of a national retail business. Below is a synopsis of the story I told. It flows as follows:

1. We don’t really know generative AI as well as we think.
2. There is a series of use cases where it can be useful — here are some examples.
3. If I were running a traditional business, I’d probably opt for this approach (scroll to find out!!1!)

We, as eager participants in the VC/tech ecosystems, have been banging on about GPT, generative AI, LLMs, and the consequent inevitable demise of humanity for quite some time. It’s easy to get excited by a concept, by a vision, and by the promise of technology that sells on the pretty strong value of automating away all the bits of our jobs that we find tiresome. Translating that excitement about the future into a tangible value for today (and for a pretty no-nonsense audience) was pretty edifying.

We all have a sound bite we regurgitate to our non-techy friends when we’re trying to seem pioneering and demonstrate the hipsterism of ‘knowing about LLMs before they were cool’, but I bet the number of people who could talk convincingly about today, tomorrow, and the next 6 months forms a smaller cohort.

It sort of goes without saying that I didn’t know nearly as much about this topic as I thought I did. I assumed I would put finger to keyboard (pen to paper sounds better but…come on), and inspiration and genius would simply cascade. Turns out, this stuff is pretty complex. Who knew.

I won’t bore you with my infantile take on LLMs, or the copious homework I did on generative AI, but I will share what I expressed to this excellent traditional retail business that is looking for an unfair advantage to get (in their case, stay) ahead.

Straight away, I defer to McKinsey’s excellent analysis. The below diagram helps demonstrate the usefulness of generative AI to the various units that make up a typical modern business.

For the business in question, the use cases I thought would be most relevant were:

1. Chatbots for the eCom site. The business handles a high volume of online customers, but is not set up like pure-play eCom retailers would be.
2. Product description generation (and enhancement), SEO-optimised as a default.
3. Sentiment analysis for the extremely high volume of communications the business has with a very discerning clientele.
4. Content creation at scale.
5. Translation of copy for international audiences (both customers and suppliers exist across Europe).

I also suggested that designers might use Midjourney for inspiration. The business is a luxury bathroom retailer, and it could be quite powerful to translate a customer’s verbal expression into a visual representation within a matter of seconds. Midjourney had a crack at:

A) “A photorealistic image of a shower enclosure in flecked white marble”

B) “A photorealistic 8k image of a luxury bathroom”

..and…

C) “A futuristic bathroom design incorporating bamboo and lots of mirrored glass”

OK, these aren’t CAD designs, but the first brush stroke is often the hardest…

But as I compiled the presentation for this business, I knew — in my heart of hearts — that these suggestions wouldn’t really resonate. I knew from my own experience with ChatGPT that the initial excitement would soon dissipate in the light of the reality of a working day, and that ‘getting good at ChatGPT’ would sink down the list of priorities as higher-urgency tasks demanded my attention.

It is critical and maybe a little cynical of me to suggest this, but I also think it’s realistic: for most businesses, ChatGPT remains a bit of a gimmick.

That is not to say, though, that generative AI is still ‘a thing of the future’.

The process of exploring ChatGPT and its underlying model was useful. It focused my thinking to recognise ChatGPT’s limitations, and forced me to see the technology through a very dispassionate lens (rationality is anathema to someone whose literal job is to get excited about an unproven future). The conclusion that I arrived at was that for the overwhelming majority of businesses, ChatGPT is unlikely — in its current iteration — totally to transform any modi operandi.

For this specific business, I selected a few application software tools that restricted the breadth of ChatGPT (and other underlying technologies) to laser-focused, clear-ROI use cases. In their case, I thought CopySmith, Longshot, and WriteSonic would be worth exploring properly, and I have challenged them to identify a few key bottlenecks in the business where they are identifiably resource-constrained, where the superhuman compute power of generative AI might usefully be leveraged.

My parting advice for their business, and indeed for any, was this: we have still barely hit the innovation inflexion point. So far, we are holding a damp match near the fuse thread of an atomic bomb of ideation and creation, the explosion of which will be visible from space. I am already a jittery nervous wreck trying to keep up with the number of businesses hitching their wagon to the generative AI locomotive, so finding trusted voices to filter signal from noise is essential.

The ones I trust to do a good job are:

1. Ben’s Bites — this is an excellent daily newsletter that combines editorial with a breakdown of cool AI tools.
2. Deep Learning Weekly — a newsletter that conducts thorough examinations of the AI/ML landscape, and writes about them really, really well. No prizes for guessing its regularity.
3. TLDR AI — a daily newsletter about AI, ML, and data science.
4. Shamelessly, I also plugged SuperSeed’s excellent newsletter.

The final thing I suggested was appointing someone to keep an eye on this world. A ‘Special Projects Lead’; a ‘Chief Geek’; a ‘Head of Future’. The tools that harness the power of this technological advancement, and apply them in surgically precise, high-ROI environments, will give such an ungodly competitive advantage to their customers that laggards will simply wilt and perish.

Oh, and if you’re a business building one of those tools, we’d like to speak.

If you strongly disagree with any of the above, great! I’d love to be schooled, and I’d love to hear your thoughts. Please feel free to comment, mock, dismantle, and dissect.

I’d especially love to hear from anyone who reckons they’ve got a good grasp on the generative AI market: who’s worth following, which are the best tools so far, and anything else that is worth knowing! Inbox always open.