And talking of listening.. you can hear this post out loud via the JEN AI podcast!
// What is AI actually useful for?
This week’s Ultimate AI Challenge comes from legit media professional Caroline Crampton, who throws down the gauntlet to the AI hype crew:
“I have a challenge for jen_ai - find something I can use literally any AI tool for. I'm not averse to it being useful, I've just yet to find literally anything it doesn't make take longer and be shittier.
It's fun to play around with, but at least for me, not actually usable.
What does it think it’s for?“
Caz, I *very briefly* considered AI-voice-cloning your voice from your podcasts to read out your question - as a fun surprise! - but then I remembered that would be.. unethical and terrifying. And as Google used to say - Don’t Be Evil.
To me, Jenni Munroe - a former Google DeepMinder and holder of a slightly out-of-date MSc degree in Computing Science specialising in AI - a machine possesses some artificial intelligence if it can carry out tasks which used to require human intelligence, like complex problem-solving and learning.
Navigation is a classic use of AI, and we’ve progressed from solving shortest-path math problems to satnav & Google Maps to self-driving cars. So that’s one of my fave use-cases of useful AI. But I’m assuming Caz meant generative AI, or the ability to create content like text, images, music, code or even molecules. She confirmed she doesn’t write her podcast transcripts by hand (although she mentions her built-in AI transcript tool seems to be most accurate on white-American-dude voices). Auto-transcripts aside, what *is* GenAI good for?
Well, for one - BS-ing. By which I mean “BrainStorming”, obviously.
(Side note: Complaints about AI learning the very human skill of “making up facts that sound semi-plausible”, aka bullshitting or “hallucinating” as the AI community call it, sounds a bit like double standards to me. But that’s a topic for another day.)
I asked the Google Gemini chatbot to showcase what it’s useful for, and brainstorming was its very first suggestion.
Gemini also thinks it’s great at explaining, summarising, editing, teaching, translation, research, emails, images, composing and streamlining tasks. We’ll touch on some of these, but more another time.
On the JEN_AI project each week we pick a hack-challenge and use AI, tech, culture and media to learn the latest in AI, tech, culture and media.
Today it’s JEN_AI Challenge #006: Brianstorming
// AI for Brainstorming
Gemini & I aren’t the only ones who like the idea of AI for AI-deation. Caz came across this Substack blog post from Dan Davies on our new “AI teddy bears”:
The way that people seem to be using it is more like the technology that used to be called “talking to a pillow”.
We’ve created a cybernetic teddy bear;
something that helps to sustain an illusion of conversation that people can use in order to facilitate the well-known psychological fact that putting your thoughts into words and trying to explain them to someone else is a good way to think and have ideas.”
Caz tells me that lots of recording studios have teddy bears / spoons with faces on for this purpose.
In programming there is a concept of “rubber duckie debugging”, where you solve a problem by talking it through aloud, speaking to a near-silent supportive listener who may as well be an inanimate bath toy. It works. Surprisingly well.
I would argue that the AI does more than the teddy bear. Maybe not enough to justify the $billions price tag for this use case alone, but it is more of an all-purpose advisor than a pillow.
// Yes, and…
My own brain went through a bit of an ideas explosion recently. After 5 years of deeply disrupted slumber during pregnancy and then as a parent of 3 young kids, I somehow, finally, got a few good nights’ sleep. Unfortunately my brain took this as a signal to flicker back to life, like the lights on a cobweb-covered arcade game in an long-since-abandoned arcade, and I’ve not been able to turn it off since.
The Large Language Models (or chatbots) have become my middle-of-the-night partners-in-crime for doing these JEN_AI challenges and creating crazy-ass ideas. In my experience, the LLMs’ practical suggestions and clarifying questions absolutely smash through writers’ block and task paralysis. Their constructive critiques nudge you onwards. They encourage. They point you in a promising direction. And even when they get (for instance) the content wrong but maybe the structure right, I often find it perversely easier to make progress by correcting something than by tackling a blank page (and yes, I have used that technique at work to get people to complete things!).
In improv comedy there is a guiding rule that “there are no bad ideas”. A concept of not saying “No” but instead “Yes, and..” - where if somebody says or does something daft the other performers channel it forwards, adding good stuff on top, and recasting that blip as a spark that builds into something better. Similarly, the LLMs take a single stupid thought and absolutely run with it. Here are a couple of examples (the vid gives a better sense of the back-and-forth, converging and diverging idea “refinement”):
Nobody has time to read or study, but we can doomscroll moviestar press-junkets on TikTok for hours. How about JEN_AI uses addictive algorithms for good and brings you Celebducation? Ryan Reynolds as multiple Deadpools explaining Relativity Theory’s twins paradox? Or Ariana Grande bigging up the science of the very small?
[not pictured, cos space/time] How about a financial-fitness app, Wealthdrip? Part of a broader life-optimisation brand? Those dodgy business ideas that “That One Friend” (now me, apparently) is forever spouting now have a path from napkin to IPO. Is this a good thing? Is it the best path? Probably not, but let’s give it a go.
// How do the Big Bots compare?
Aside from Google’s Gemini, I also challenged OpenAI’s ChatGPT (above) and “ex-OpenAI-team” Anthropic’s Claude to brainstorm what we can usefully do with GenAI. Here is Claude’s effort:
Hey! This looks like an opportunity for MORE AI HYPE! 🥳 Let’s go!
Bonus points for making mind maps and trying to write me a better blog post. A lot of people like Claude’s more human tone of voice, but wish it had an option to search the web in real time.
// Gemini Live: real-talk
Gemini has an out-loud voice-chat mode called Gemini Live. I was crying out for this feature when I was using ChatGPT heavily at the start of this JEN_AI project. It has a slightly stilted conversation flow for me at the moment and it sometimes just ignores me completely (probably overwhelmed), but it’s getting somewhere and has been great when my hands are full and I don’t want to stare at a screen. It transcribes the conversation for reference, v helpful. Hopefully this feature will get smoother soon.
Edit: yeah, ChatGPT caught up.
// The importance of Context vs. Prompts
You hear a lot about “prompt engineering” - I.e. getting the knack of writing a really good guided question for the chatbot so it gives you the output you actually want. But I don’t hear as much about the significance of context. Because I’d started off this JEN_AI project using primarily ChatGPT, it knew my vision, some challenges I wanted to explore, values I hold, motivations, points on my life journey, my preferred tone and love of lame wordplay.. and more. So now it can give me helpful responses even if I just throw a badly-written half-sentence at it. An example: I checked my work emails and realised there was a deadline that day to submit a talk for a conference. I had very little time so I just quickly pasted in the email and asked ChatGPT to suggest a talk topic for me. It already knew what I wanted to say. A few quick iterations and bam.
*Teaching the bots context is as valuable as writing a good question prompt.*
So, the question of whether Gemini vs. ChatGPT vs. Claude vs. Joe LLM is best for you will also come down to which one you have used (and therefore personalised) the most.
// Privacy.
A valid concern: what are the Big Bads at Big Tech doing with all this info you share? Well, I need to do a bit more digging to truly get to the bottom of that important and terrifying question. But in the meantime it’s worth making a few defensive manoeuvres, like not sharing things you don’t want them to know, and checking and adjusting your privacy settings, e.g. to not feed your ideas back to train models in ChatGPT. Last I checked Gemini is slightly better on this kind of privacy, it’s in by default.
// How did we do?
So how did we do on the challenge of brainstorming useful uses of GenAI? Is the Ultimate AI Challenge turning into more of an Ultimate AI Quest for the elusive Holy Gr-AI-l? (Another joke fail. Sorry folks.) Probably we should crack open some code at some point, surely auto-writing software has some utility? A worthwhile function? A helpful application? Nope?
Let me know! Just like an AI teddy bear, I love to hear your ideas.
// Bonus Material: JEN_AI Playlist
I ran out of time and space this week (that’s school holidays for you) so I had to scale back my ambitions for this week’s podpost. I was desperate to include a Matrix-movie skit of me having the redpill/bluepill chat with an AI Morpheus followed by a primal scream kind of brainstorm rage explosion rant. Oh well. We all probably dodged a bullet there.
But I did start adding a theme song for each challenge/episode to a JEN_AI Spotify Playlist. The vibe is kinda “The Future But Retro”. This week’s theme song(s) are:
“What Was I Made For” by Billie Eilish. Applies equally to Barbies and AI.
“Brianstorming” by the Arctic Monkeys. Yes, the “ai” is intentionally backwards. Seems apt.
This playlist is almost unlistenable due to the clash of genres. Which could be a tagline for the whole JenAi project! If you meet anyone who likes both Madonna and Breakbeat Electronica let me know, we should be friends.
Lastly, here’s a quick bio for this week’s awesome virtual guest, according to ChatGPT:
“Caroline Crampton is a British writer and podcaster, author of A Body Made of Glass and The Way to the Sea. She created the award-winning podcast Shedunnit and serves as editor-in-chief of The Browser newsletter.”
A big thanks to Caz for contributing to this experiment and being our first-ever human special guest via the medium of WhatsApp.
// Get Involved!
If you, the reader-listener want to take part in the JEN_AI project, we’d especially love to hear your takes on the present and future of social media, so wing those over up to Friday 21st of Feb 2025 to make it into the “Social Mediums” episode.
Or you can always, anytime, send us your thoughts or questions about JenAI via any sociable medium.
And, as always, thankyou so much for listening.
Share this post