The robots are here and they don’t suck.
I’m taking a break from our usual programming to chat about something topical: the latest wave of generative A.I. (Don’t worry, we’ll be back next week with more endurance, running-ish writing.)
Like many, I’ve used A.I. to get meal recipes and vacation advice. But I’ve also used it to help create this newsletter. I thought it might be fun to lift the hood, so to speak, and share how I’m writing with the bots.
So here are 5 ways I’m using A.I.
1) Feedback partner
Getting and receiving feedback is the favorite part of my day job. Thinking through a design or piece of writing with other people is thrilling. It can be like improvisational jazz—human minds riffing off each other with the materials of language.
Since this newsletter is a one-person shop, why not riff with a robot? ChatGPT can bring a different lens to a piece of writing.
For example, last week’s post was a weighty exploration of religious belief and athletic performance.
When I was making edits, I asked ChatGPT to pretend it was an editor at The New York Times and make a chunk of the essay more concise.
I didn’t directly lift what it generated, but ChatGPT spotlighted places to tighten and simplify for impact and ease of reading.
Shapeshifting editor. I could also ask ChatGPT to mimic the house style of other media outlets (The Atlantic, The Economist, Axios) to provide different editorial frameworks.
ChatGPT can also toss ideas on the table when I’m feeling stuck.
Recently I worked with the chatbot to create a name for some news digests to distinguish them from longer essays. Here’s what it came up with after some refinement:
I didn’t use these titles, but they spurred my thinking and kept me concise. I landed on “Straights and Curves,” like of a racing track. Thinking alongside the chatbot helped get me there.
2) Eye candy
I’m a fan of OpenAI’s imaging software Dall•E 2. The images are fun and easy to make. Just enter a description of an image and the A.I. tries to create it.
The header image above and the one below were created from the prompt: “An oil painting by Matisse of a robot typing on a computer.”
Dall•E 2 is also a useful design tool.
Reading on an electronic device is different from reading on print: your attention is fragmented and more precarious.
Images break up streams of text with some visual flair, keeping you engaged with the text as you scroll. The ease of A.I. means more images to play around with layout and structure.
The ethics here are blurry. You could argue that by using A.I. art tools, I’m stealing work from real photographers and artists. I’m employing digital scabs, so to speak.
It’s not hypothetical. Artists are suing the A.I. art generators Midjourney and Stable Diffusion for scraping their work.
Sadly, paying for photography to jazz up this little newsletter isn’t financially feasible. And I’m a terrible photographer. So it’s A.I. images or no images... I prefer images.
And yet I’m sympathetic to the working artists who argue their intellectual property has been stolen by the image models. What are your thoughts?
3) Assistant for mundane tasks
I’m not great at writing catchy copy for social media, adverts, or landing pages. Such writing doesn’t bring me much joy either.
So why beat my head against the wall? Here are promotional tweets that ChatGPT pumped out for this newsletter:
Is this agency-level? No. Is it better than spending an hour of my life to toss stuff on Google Ads or Twitter for advert clicks? Absolutely.
Now served with swag
Get a Breakfast Club hat and stickers at our modestly-priced Swag Shop.
4) Definitions and spearfishing
ChatGPT is great for quick explanations. For a recent letter, I wanted to check my understanding of the NCAA’s Name, Image, and Likeness policy. Chat GPT provided a concise summa:
This is quicker than Google search, which is a more navigational form of discovery where you skim through multiple results. ChatGPT is more like spearfishing. Plug in a query, get back a contained response.
5) Inspiration instigator
When brainstorming, ChatGPT can churn out metaphors to spark ideas. For my day job, I asked the chatbot to create metaphors related to privacy:
The idea of privacy as a “cloak” was new for me, especially as a verb in relationship to the self.
It made me wonder how I might re-conceptualize privacy less as something to defend from public intrusion, and more as a protective tool like the cloaking devices Klingon warships use in Star Trek.
See? Ideas already sparking.
Here are places where I’ve found ChatGPT falls short:
As a proofreader. The bot doesn’t make many typos, but it’s not great at catching stylistic errors or grammatical infelicities in your own writing.
Generating long-form arguments. After headlines about ChatGPT passing bar and med school exams, I plugged in a few essay prompts from my TA days at Cal. It wasn’t pretty.
As a virtual assistant. I’d love to unload work email and Slack messages to a chatbot. While ChatGPT can create stock email responses for set-piece situations like “asking for an extension to my paternity leave” or “politely following up about my last message”, human communication is rarely so boilerplate. It’s complex, nuanced, and cued to hyper-specific conditions. The chatbot can’t handle that yet without significant inputs from a human.
Dive deeper. There’s so much to say about A.I.’s potential impact on society and the economy. Here are reads and listens I’ve enjoyed:
Placing A.I. into the history of how people relate to knowledge (Economist)
Ezra Klein’s skeptical take (NY Times)
Casey Newton and Kevin Roose on A.I.’s effect on jobs. (Hard Fork)
Roose’s unsettling conversation with Sydney, Bing’s chatbot (NY Times)
Are you using generative A.I. in your life? If so, I’d be curious to learn how.
Weekly run
Breakfast Club meets every Thursday for an 8-mile run:
When and Where: 6:30am at Lake Temescal in Oakland, CA
Pace: ~7:00 to 7:40 pace with a few hundred feet of climbing
Tweets of the week
Parting thought
“Even if [machines] did many things as well as or, possibly, better than anyone of us, they would infallibly fail in others. Thus one would discover that they did not act on the basis of knowledge, but merely as a result of the disposition of their organs. For whereas reason is a universal instrument that can be used in all kinds of situations, these [machine] organs need a specific disposition for every particular action.”
- René Descartes
That’s all for this week. Thanks for subscribing.
Great insights Sam. I've had mixed results with ChatGPT. It's good at taking something you feed in and returning a rephrased or copyedited version, or returning 5 different ways to phrase something, however you will invariably have to edit what it puts out. A big deficiency is its inability to learn from a "conversation", which it keeps in its record. Supposedly this is one of the great things about AI, its ability to be trainable. A tool I am aware of at https://www.grammarly.com/grammarlygo is an interesting AI tool for writing, check it out.
I have a lot to learn about how to train it (say, to generate content in the PayPal voice). In a recent conversation with ChatGPT, I directly asked it to never apologize, as it is very prone to do with tiresome boilerplate apologies. it continues to apologize, and even apologizes for apologizing. My other huge concern is bias. As with a search engine, it can be manipulated by its trainers to generate selected responses, which can easily tread into the political spectrum. For instance, I asked it: Why is Christopher Columbus considered to be great? It replied: While he is often celebrated as a great explorer and discoverer, there are also those who criticize him for his treatment of indigenous peoples and for his role in initiating the colonization of the Americas. On the positive side, Columbus is credited with...". This is clearly a political statement vs an answer to my question. It spells out the negative first, then talks about why he might be considered great. There are other AI chatbots for content being built as this space rapidly evolves. It will all get better.