Typos as a feature
How do I prove to you that I’m a person? Do I need to spell strawberry correctly or maybe wrong? Tell you to take your car to a carwash? Draw a pelican riding a bicycle? Do I need to stop using dashes? Should I start cursing? I’m going to curse profusely in this. What about typos? Should I misspell stuff? Should I write gibberish?
LLMs are so good at churning out tokens. Why doesn’t everyone have a novel? Why doesn’t everyone have a textbook to their name? Are people not prompting enough? Are we prompting enough to justify the datacenters?
Everyone doesn’t have a textbook because there are people with credentials and hard fought “I’m that fucking person that knows this shit” behind great textbooks. “I spent ten years in a basement grinding this out so, yeah, I’m going to write a book about it”. Unique ideas, stories, explanations come from people.
The person grinding out ten years in a basement so an LLM can regurgitate their knowledge to people who wont take the time to read it is not a feature. Imagine trying to learn Italian because you’re taking a job in Italy and cheating to get a higher duolingo score. You’ll be able to save yourself with google translate, but you’ll lose the beauty of communicating with someone in their native language. The air pods pro max can not replicate this for you.
People are making browser plugins just to identify LLM text. Its not that LLMs use inhuman words or conventions… those would be easy to recognize… but because people want to know when they are being LLM’d, and its exhausting to constantly have your antennae up. We used to be able to read things and generally assume that someone, somewhere, had taken the time to convey it well or copy it from someone who did.
There is some argument here like: “Well that just means LLMs are fucking great, people struggle to distinguish between them and actual human writers” and these people are missing the entire point. The point is that people want PEOPLE. People want people to listen to them when they are in hospice. People want people to play chess against. People want people to create art. And yes, people want people to write words to them and not get half way through an article and realize you’re basically ingesting a linkedin post on another platform.
If the communication is nothing more than bureaucracy and both sides are just pissing tokens back and forth maybe that’s an excellent sign that those comms are redundant or need to be altered in some way! I’m not LLM-shaming or AI-Bashing, they are useful, I use LLMs, but “should I be writing this?” is currently an underrepresented question.
None of this is about putting down words. Writing words is nothing. I can write words words words words, ufcking gibberish. It’s about the thought and intention behind the words. People want to know if an LLM wrote the email because the data, the opinions, the phrasing builds confidence that the author knows what the FUCK they are talking about. We want authors that know what the fuck they are talking about and have invested their time to produce something. Just as the reader is about to invest THEIR TIME to consume it. If someone takes 2 minutes to curate something with an LLM, then I’m also going to take 2 mins to ingest it (most likely via LLM) and that is fine. The asymmetry occurs when someone takes 2 mins to curate something with an LLM and someone takes an hour to consume it. That is ass backwards.
Our writing, data story telling, presenting, its all to convey concepts to other people succinctly in a way they can grasp.
Preparations for a presentation take longer than the presentation itself. When someone has spent time honing their skills as a presenter, you can tell. Hard fought battles with public speaking engagements; Zoom calls where they talked for 20 minutes and nobody speaks or even asks a question afterwards. That shit sucks, but you wont know until you eat sand.
A picture is worth a 1000 words and your written opinion is worth your reputation. Earn it.