Today’s pre-Labor Day weekend piece is my follow-up to Noah’s “On AI and Writing” that resonated with so many of you. It’s only fitting to be introspective about our craft on the eve of a holiday honoring the dignity of human work.
It’s been a big week on the BRXND event front with 30+ new registrants in the last couple of days. We’d love to see you at the Times Center on Nov. 5— if you have any questions, drop me a line at mike@brxnd.ai.
Since Noah published “On AI and Writing” two weeks ago, the discourse around AI as a “legitimate” medium for communicating original ideas has only intensified. Tim O’Reilly wrote a contrarian defense of writing with AI arguing it is “a field of possibility from which humans can and will summon up and communicate insights, truth, beauty.” Stanley Druckenmiller proudly published a Wall Street Journal op-ed with AI and was resoundingly backed by the paper’s editors. On the flipside, Derek Thompson shipped a phenomenal interview with Pangram founder Max Spero pleading that “we can’t let AI writing take over the internet.”
Suffice to say, for folks like me who attach a significant amount of their professional identity to being “writers,” it’s been an especially intense two weeks where it’s become ultra clear that the genie isn’t going back in the bottle.
Like Noah, I started my career as a journalist and maintained an active second career as a ghostwriter, blogger and newsletter author throughout my decade or so working in technology companies. No matter what I was doing in my day job at the time, any writing project was always the “hardest” work I was doing. Staring down the barrel of the blank page burned more intellectual calories than anything else I was working on.
This dynamic has only gotten more intense in the age of AI. LLMs are much better than I am at “low-caloric” writing tasks, leaving only the arduous work of thinking in their wake. Therein lies the agony and the ecstasy of this moment for anyone who writes words online for a living. The job is harder than ever but the work that remains is the most gratifying aspect of why we do this.
Building on Noah’s piece, here are a few thoughts I have for writers trying to future proof their craft and understand their raison d’etre as AI eats the world.
While I agree with Noah that the more powerful models are getting worse at writing in many respects, I believe they have crossed the threshold of being better at the “view from nowhere” style of summation analysis than even great human writers. For an omniscient take on most subjects, AI-generated text does the job at hand. You’ll probably even learn to love some AI-generated work.
So what remains scarce when the majority of what has been written for the web is done well by LLMs? Insights derived from true shoe-leather reporting, analysis drawn from learned and felt experience and things that make people laugh. Writing that doesn’t check one of these boxes will be owned by a machine.The original promise of the internet is that it was supposed to free writing from the rigid form-factors of newsprint and allow unbounded creative expression. In practice, so much writing for the last two decades has been self-censored to fit in the SEO box. We are increasingly being freed from this.
However, the tough thing for writers adjusting to this new world is that before LLMs, there was an enormous amount of demand for “view from nowhere” takes optimized for SEO. The market made us into one thing and is now aggressively demanding another.
In a similar vein, Sam Parr has a really good “barbell” theory of internet writing that the best posts are either <500 words or >2K word long essays. His advice is to avoid the messy middle length that defines much internet writing. This is more true now than ever before.
It is also really hard! The ~1,000 word messy middle is the easiest form factor to make an argument in and generally where the narrative wind blows me. This is largely due to the fact that for two decades, 800-1,200ish words was the standard length of piece commissioned by editors for op-eds and many manners of freelance writing.
I am deeply struggling to reframe my craft around 400-500 word quips, supplemented by the occasional piece d’resistance type essay. Note the length of this piece! But I believe this to be a near existential thing to get right.I agree with Noah’s take that the best writing is fundamentally anti-consensus. But this presents a conundrum for writers who have to ship content on a consistent cadence. The reality is that you just aren’t going to have that many good, contrarian takes! So how else can you consistently write pieces that can’t be replicated by AI?
Many of the best pieces of internet writing are something that an industry, subculture or at times an entire society is deeply feeling but lacks the precise eloquence or courage to coherently say out loud. I believe this “finger on the pulse” style of writing will be safely LLM-proof for a long time and will become even more valuable as emotionless AI-generated copy eats the web. For now, models simply cannot “feel” vibes in real time at nearly the level people can.Lots of the best internet writing is going to look increasingly rough around the edges. I believe we’ll see a style emerge that looks less like traditional essays and more like “jagged collections of bangers” thrown together without the usual transition paragraphs.
This will be a response to both reader demand for brevity and a counterpositioning against the tendency of LLMs to create flowy, overly cohesive narratives. It will be weird for a while but ultimately liberating for readers and writers alike.
For my money, Matt Levine is the consummate example of the most AI-proof writer on the web. If you want to have a good laugh, try prompting a model to write an essay on a topic you know well “in the style of Matt L’s Money stuff.” AI is horrifically bad at capturing Levine’s je ne sais quoi of wit backed by a couple decades in the boiler room.
Honorable mention here goes to Sam Kriss as a sheer master of prose that operates on a level that simply can not be emulated.
To steal from Picasso, LLMs are going to be better than you at “learning the rules like a pro.” They will not be better at breaking them like an artist.
If you have any questions, please be in touch. As always, thanks for reading.
— Mike
Great one!
For other “AI could never” writing, see David Roth’s writing in Defector