They say if you want to understand something, teach it.
Last week some colleagues and I got the chance to do just that at our first AI in Action workshop, which we held at Trust Square in Zurich.
This was a pilot of a series that, under the aegis of our newly formed Applied AI Collective, we had been developing to showcase our own experiences with generative AI in business contexts, and help others start their journey.
(And in fact it was with the help of some nifty AI tools that we were able to develop, market and run the workshop in less than three weeks).
Below is an account of the day and some of my personal takeaways.
1. It’s a brave new world for all, regardless of background or technical expertise
We had managed to get a diverse crowd of around 20 people who were willing to pay the CHF 99.00 francs we were charging for this pilot (it included lunch).
There were professionals from global banks and fintechs, startup founders, technologists, students, people transitioning between jobs, even the ex-owner of an Italian football team.
I had been concerned that the more technically minded types, whom I assumed would have been familiar with these tools already, might be bored by our intro-level demos. The contrary was true. Regardless of previous experience, everyone seemed to find something to intrigue them.
There was a blockchain founder on a budget who saw the website, video and graphics generation tools we showed and realised he could finally have the quality marketing he was dreaming of despite his limited budget.
The co-owner of an innovation company said she’d been using ChatGPT for months and found it really bad, until she realised in the workshop that she was prompting it wrong. After five minutes, she said, she was hooked.
Another founder was bowled over by AgentGPT, the autonomous AI agent, and used it to start generating new ideas and products for his trading platform. (I happened to run into him about a week later and he told me he hadn’t been away from his computer since the workshop, he was just drunk on all the ideas the machine was giving him.)
The takeaway: there is something in this for everyone, the trick is to dive in and get exposed.
2. Use cases seem endless — and probably are
This was billed as a practitioner’s workshop. And for good reason.
None of us are machine learning experts or data scientists. We are however all business and creative professionals, with backgrounds in fields like strategy, innovation, creativity, communications or banking. And we all had been heavy users of generative AI since it first burst into the mainstream last January, using it daily in our own work.
So we naturally focused on practical demonstrations. (For a full list of the tools we discussed, see the end of the post.)
To give a few examples:
Mark showed a storytelling app he had built on his phone which could take a few ideas (a person’s name, an animal, a place) and create a short story and visual from it in a few minutes.
He also showed how he used AI to analyse a spreadsheet containing names of conference participants, match those with common interests, and then write personalised introduction messages connecting them.
Iwona took us through image generation, showing how she used AI to create visuals for her business (and for the workshop).
I demonstrated a number of projects where I had used AI for communications and content, going through the chain from brainstorming and outlining, to research (summarising material, internet research) to the actual writing.
Then I showed a number of tools to turn these ideas into websites, almost instantaneously.
Watching people experiencing these tools for the first time reminded me of my first exposure: there is initial disbelief that this is real; that a machine can seemingly think, seemingly be so human. And then it dawns on you: because these things run on natural language, there is at least in theory almost nothing you cannot at least ask it to do.
The takeaway: there is no part of our professional lives that these tools and this technology won’t affect in one way or the other.
3. It’s far from perfect, but these are early days
In my presentation I made a point to show not just what tools like ChatGPT can do, but where they go wrong.
In my own work I had experienced a number of what are known as generative AI “hallucinations”: a book summary that contained invented chapters; references — complete with author, title, date and publication — to articles that had never been written.
It was important that others see this too. Because one of the main dangers of generative AI, at least as it is today, is that it isn’t fully trustworthy but comes across as extremely confident in itself, it can lead the unwary dangerously astray.
After the jaw-dropping moments of experiencing generative AI magic and understanding what it can, I think observing its mistakes may have been the second most impactful thing we did from participants’ viewpoint.
But as Mark above all was keen to point out, time and again, these are early days. The below pizza picture, which shows Midjourney getting better at conjuring a pizza image over time based on the same set of instructions, shows just how fast these things can improve. Anyone who has seen the massive improvement between GPT 3.5 and GPT4 can only wonder what GPT5 will bring.
It would be foolish not to think that today’s limitations won’t be surpassed, and soon. What this means is anyone’s guess right now, but I sensed it gave people at the workshop a lot to think about.
The takeaway: for all their glitz, these tools need to be used with caution.
4. Generative AI has the perfect user interface — people will need to get used to that
After our demos we turned participants loose in groups to develop their own projects.
A few things stood out for me.
One was the amount of experimentation: business ideas and strategies, marketing materials, new kinds of research tools. I have rarely seen workshop groups so totally absorbed.
Another was seeing how people needed to get used to interacting with a machine by simply talking to it. We’ve seen this in science fiction forever, but it is another thing to experience it live.
And to think what that means. At one point someone asked me for a tip and I said, “Don’t ask me — ask GPT”. To which he responded: “Of course. I keep forgetting I can talk to it.”
The takeaway: the fact that generative AI tools seemingly understand us is oddly strange. That they can explain themselves does take some getting used to. So does the fact that you can use one tool to run the other, like asking ChatGPT to write Midjourney prompts.
5. Summing up
To close the day, Stéphanie and Celine led a discussion on key takeaways.
As I have tried to show already, there was interest, excitement, and no end of ideas among the participants.
It would only be fair to point out that there was a fair amount of anxiety too: what does it mean for my job, for privacy, for the rights of the writers, artists, thinkers and other creators upon which the large language models that power generative AI are based.
These are all questions that this technology raises, and that I will be addressing in future posts.
In the meantime, we are looking forward to putting on further workshops.
Would you like to learn more? As mentioned above, I am running a short series of free 1-hour AI Practitioner’s Intro webinars. Sign up here.
About the Applied AI Collective
The Applied AI Collective is a community of early adopters of generative AI-driven tools. Our intention is to build a practitioner’s hub where we can help others gain the AI skills they need to innovate, be more creative and productive, and grow their businesses. We are:
Our workshop was about much more than ChatGPT. Here is a list of the tools we either demoed or discussed:
- Chat GPT: https://chat.openai.com/
- Bing: https://www.bing.com/
- Bard: https://bard.google.com/ (not yet available in Switzerland)
- Stable Diffusion: https://stablediffusionweb.com/
- Midjourney:Go to https://www.midjourney.com/ and click on “Join the Beta”. You will be brought to a Discord server. You interact with Midjourney via Discord.
- DallE: https://openai.com/dall-e/
- RunwayML: https://runwayml.com/
- Perplexity: http://perplexity.ai/
- Consensus: http://consensus.app/
- BundleIQ: https://app.bundleiq.com/
- Summarize.tech: http://Summarize.tech
- ChatGPT Splitter: https://chatgptsplitter.com/
- Otter.ai: https://otter.ai/
Design, presentation, website building, video
- Microsoft Designer: https://designer.microsoft.com/
- 11Labs: https://beta.elevenlabs.io/
- 10Web: https://10web.io/
- Durable: https://durable.co/ai-website-builder
- Mixo: http://mixo.io/
- Beautiful.ai: https://www.beautiful.ai/
- Designs.ai: https://designs.ai/
- DeepBrain AI: https://www.deepbrain.io/
- Glia studios: https://www.gliacloud.com/en/
- Synthesia: https://www.synthesia.io/