7(ish) Questions with Jenny Nicholson
LinkedIn is a weird place, made even weirder with AI. There are so many people SO CERTAIN about how things work before they even come out, tossing out best practices for using things they have never touched, so many people clamoring for attention and adding nothing but noise and confusion.
And then there is Jenny Nicholson. She posts about her experiments and experiences. It’s a proactive, positive version of fuck around and find out that I absolutely adore. When GPTs came out, she not only figured out how to make one that worked, she started a project to make one a day for the entire month of January, really figuring out what they could do, and where they struggled. So I had a few questions…
Do you remember your first interaction with generative AI? What did you make, and how did you feel about it?
My very first intro to generative AI was in 2021 with GPT 3. It was just a completion engine then where you started a sentence or a passage, then hit "enter" and it picked up from there. I made a dumb poem, which even then felt like magic. There was an unhinged quality to 2021 GPT and it was definitely not "safe" at all. But tbh I kind of miss it. I remember feeling like I was tapping into the collective subconscious. You can still get to that place with today's LLMs, but it takes a lot of careful prompting.
I’ve realized that I have very different emotional reactions to different LLMs, and find myself going to one for some stuff, and another for other things, and I’m kind of starting to feel like a crazy cat lady, but for AIs. Do you have a “favorite” LLM, and what kind of “relationship” do you have with them?
Yes I relate to this so much! I often find myself bouncing around between Claude Opus, GPT-4, Gemini Ultra, and Gemini 1.5 Pro. In terms of overall vibe, Gemini Ultra is probably my favorite. But when it comes to complex tasks or really advanced prompt engineering, Claude Opus is my go-to every time. I still do a ton of work in ChatGPT every day, mainly because I do a lot of GPT building for clients and nobody has built out the full tool ecosystem as well as OpenAI. And I def think people are sleeping on Gemini 1.5. A million tokens of context is A LOT and there's a lot you can do with that much memory.
Let’s talk a bit about GPTs. You did a project where you created a GPT a day for a whole month, and they really – I think – helped people understand their potential. You created gpts that responded with visuals, scripts, ones that are made for conversational mode, and some that teach you something about yourself, or even put you in front of a panel of AI judges. One of the most fun things for me was how many different uses you found for them, and I loved the unexpected twist of the “imagine a sheep” GPT you made.
Do you have a favorite?
First off, I have to give a shoutout to Allister Hercus, my partner in the Lab31 project. As great as AI can be, nothing replaces a human partner when you want to make things happen. Instant Party is the one I find myself using the most (which I did not expect). It's been great for those random long weekends where I underplanned and could have ended up plopping my 10 year old in front of the TV. But now, it's like, wanna have a party? And no matter what day it is, we've got something to celebrate. The Project Runway GPT has been super useful too. After four years of working from home, I need all the fashion help I can get.
Were there any disappointments?
The biggest disappointment for me was the AMA GPT. I'd spent all day getting the model to faithfully capture the voice of just about anyone at, at 10:15pm, OpenAI literally released a model update that nerfed the whole thing.
Where do you see the potential of GPTs heading when it comes to brands, or just in general?
The trickiest thing about GPTs in specific is that they're only accessible if you have a ChatGPT+ account. And I am continually BOGGLED by how few executives in marketing have dropped the $20/mo to subscribe. (Seriously, y'all, it's $20. DO IT!) If OpenAI continues their market dominance and more people get paid accounts, it'll be an incredible way to connect with customers. Especially once the memory feature is properly released. In the meantime, I see custom GPTs as a great way for brands to test out new product features and interactions without a huge monetary investment or touching their core site. I just launched a custom GPT with Zola that helps couples more fairly divide wedding planning duties, which was super fun. It even gives you a downloadable CSV with your assignments and resource links!
As for AI in general, I honestly have no idea. If anyone says they do, they're lying. In the near-to-midterm future, the focus is going to be on production at scale – getting more things out the door, more quickly. But I'm hoping that's not where brands put their focus. So much is going to change over the next few years, not just in terms of technology, but in terms of people's behavior. Where will we go for trustworthy information? What will happen to the search experience? How do we balance the ability to hyper-personalize marketing using hyper-persuasive LLMs with the responsibility to use our powers for good? I spend a lot of time thinking about this stuff and I hope every major brand is thinking on this scale, too.
Your LAB31 project was not only ambitious, but I’m guessing incredibly useful in understanding how to bend these sneaky little bastards to your will. Any tips? I find getting them not to respond in lists is really hard, and kind of “the simpler the better” with a lot of these.
First off, always focus on telling them what you want them to do, not what you don't want them to do. Also, you can define the structure of the output however you want and the model will match that. The models are trained to please you and, if ChatGPT gives you a ton of lists, you can blame a) Buzzfeed and any other listicle in the training data and b) the human reviewers who trained the model to think that “list = good”.
And yes, simpler is usually better. Two of my favorite, most personality filled GPTs I've made – Lazy LLM and Ennui Bot – both have like three sentences of instruction. I often find that, if I'm adding tons of specific information and trying to perfectly describe every step, it's a sign that I haven't framed the job to be done clearly enough.
I have a BreeGPT that I made mostly to amuse my friends and annoy my husband. Have you made one of yourself?
LOL, not sure the world is ready for TWO Jennys.
What do you think the most important thing creatives should (explore, know, understand, try – feel free to pick one) when it comes to generative AI?
First off, I understand why so many creatives are resistant to this technology. Yes, it's about protecting our livelihoods, but it's deeper than that. Creativity, art, music, writing – these have always been core to what makes us HUMAN. So seeing machines mastering them brings up some pretty gnarly questions.
But the truth is, this technology is US. I hate the term "artificial intelligence" and very much prefer "collective intelligence." Because that's what AI is – the combined knowledge, perspective, and history of all of us. From the writings of Plato to the depths of a random subreddit, the ghost in the machine is all of us. Which is weird, but also so so so lovely.
For any creative who hasn't jumped in yet, I go back to this quote from Leonardo da Vinci: "Learning is the only thing the mind never exhausts, never fears, and never regrets." You don't have to like it, but there is NO downside to learning how to work with it. (But once you get the hang of it, I think you'll like it a lot.)
