It is now possible to generate any number of words you’d like on any topic. This makes us feel smart (well, it makes me feel smart at least). We have voice conversations with our Chats and ask it to summarise the conversation into blog posts. Then, if we think it is sufficiently interesting, or if we think we have discovered new forms of quantum physics, we can share it on LinkedIn with all our professional colleagues we are trying to show off to.

That’s the exact moment LinkedIn started becoming unbearable. The easier it is to generate words… the more words we generate (Jevon’s paradox). Low-quality words are easier to generate than high-quality words, so much of the generated words we see are low quality (this is so self-evident that it seems almost stupid to write in a sentence).

By writing quality, I mean something like Novelty per Word, or the amount of information each word contributes that is both digestible and insightful.

I am guilty of this, too. If you look at my blog articles from earlier this year, you’ll see a whole pile of (yes, I use the word pile here intentionally…) articles that I generated after a creative session with coding agents on many different projects. The sessions themselves were worth writing about, but because writing is difficult, my brain took a shortcut and used AI to write the story for me.

The correct response is that we should all learn to expect higher-quality writing from each other, especially in long-form content. But this is taking some time for us all to get used to, and so in the meantime… viral slop.

Something has broken in the contract between the reader and the writer when coherent writing has become so easy to generate.

There is a core idea in cryptography called “proof of work” that underpins how cryptocurrencies work. The basic idea is that if a member of a network wants to write something into a shared history, it must deliberately do something challenging and prove it; otherwise, the shared history among the nodes becomes polluted, and no one can agree on anything. In the case of Bitcoin, this deliberately hard problem is to find the input that produces a given output under a non-invertible hash function. In the case of Writing, the proof of work used to be that it was difficult to sit at a typewriter or a keyboard and produce coherent writing. This is no longer the case.

We need Proof of Words.

I think there is a product opportunity to build tools to help human intelligence write higher-quality words.

I am still debating how I feel about using WisprFlow and Grammarly. Over the past two years, I have become an addict. I use it everywhere. In Slack, Email, Telegram, and every medium you can think of. I WisprFlow half-baked Slack responses, and then I use Grammarly to fix the grammar.

I think this is a useful tool for short-form communication. It overcomes my writer’s block when I can just start verbalising as I figure out how I want to reply. I reply to messages faster at work. I proactively send that quick message to avoid miscommunication or mismanaged expectations.

But for long-form communication, I’m not so sure. I use WisprFlow to dump raw thoughts onto an Obsidian page when I know I want to say something, but I’m not exactly sure what. I ask Claude to summarise all my thoughts and write something coherent from them. I can’t remember a time when it has ever perfectly captured what I wanted to say. I am beginning to think that it never will, despite all the AI advances that will arrive.