Automating creativity is the death of art.
The typewriter was the beginning of the end of the art of handwriting.
Factories was the beginning of the end of hand-created goods.
AI has killed the art of handwritten code (which itself automated away other forms of art).
Maybe it's like when you can just tell a picture is AI. You recognize the words you are reading, but to your brain, the way the words are put together feel wrong.
Incidentally, I was reading Harry Potter to my kids last night. While waiting for everyone to arrive, I started reading the previous chapter to those who were present. But I read the sentences backwards while using a vocal inflection of regular narration. My 6-year-old recognized that something was wrong and asked me to "stop reading funny". My 11-year-old recognized that I was reading backwards. The same 6-year-old thinks every silly photo that I show him is AI, while my older children can actually spot the AI.
Or think about reading something that might be stuff you could say in your own words but not actually in your own voice. Or maybe the opposite is a better example for you? That's what some poetry is about. The gap between voice and word.
Stylometry is the study and measurement of text, so that can go some distance to quantifying the difference between voices and can help to programmatically identify a human voice but I feel that because a voice comes from a person, the recognition is also mainly a human response.
At another level its about communication, person to person with the intention and desire to communicate something. LLM text output may seem uncanny as it's hard to emulate the intention and desire and empathy present in genuine human communication.
I would be interested in reading what poets say about LLM voices. I wonder if many LLM labs employ them?
I had Claude build me a map tool to calculate travel times from one or two points to every street in the city; not only did it do that, it added various scenarios, speed limits, usage of bus lane toggles, different visualizations based on thresholds, and a tool to scan for optimal starting points based on a configurable grid size. etc.
This overcompleteness is a mixed blessing though, in this case it was great and convenient, but if it was e.g. documentation intended for human consumption then it wouldn't be so great, and if it was a letter written to the county that they should consider another location for a fire station it'd actually be pretty ineffective.
Even if you don't think it's a problem, even if you believe that it's just different than human writing, it seems the majority of "regular" people can tell when something is LLM-written and find it distasteful or unpleasant to read, at best. That alone should be more than enough incentive for frontier labs to train this out of their state-of-the-art models, if only they could. The fact that they haven't suggests that they cannot, despite all the claimed advances in model analysis and "explainability."
That being said. Yes, I hate some LLM Voice moments too and still sometimes judge people for copy and pasting, expecting others to do the comprehension when it seems they haven't done it themselves.
Duality of man.
Look at programming. What's happening there is we are having agents produce 10-100x the code to where people give up on reading it and just hit OK. Look at math. Plenty of people there want to learn from the LLMs and all these crazy new math proofs. But those are the people who will probably be out of a job soon because the money in math is in producing proofs, and that's now an LLM-exclusive job.
I think LLMs are weird in that the first order effects are those of an incredible and amazing tool that anyone can learn things from, but the second order effects are going to be the exact opposite - people giving up on bothering with ideas. Even at my job, which is one of the few that has not gone crazy with AI, anytime I ask anyone a question, all I hear is "have you asked Claude?". There's no communication anymore, no pushback, no struggle. I worry that one day we'll look around and realize we've built an entirely separate second civilization of machines and robots that acts more like a neighboring country than a tool we can use to build humanity up.
-Just my rambling thoughts, not refined by an AI.
I worry that--while they can be a strong fuzzy search tool-- their teaching ability falters because teaching involves possessing a theory-of-mind for the student and trying to figure out what will change how they think. A good teacher doesn't just present you with relevant facts, they try to figure out how you're reaching conclusions.
Note that I'm excluding propoganda-style "teaching" here, where a model just insists you're wrong and that all the cool kids are doing it etc.
(I have to admit, I truly do like some of the music, so feel free to scorn me. If you want to be infested with a vicious ear-worm check this out: https://www.youtube.com/watch?v=zONGVaONKe8&list=RDzONGVaONK...).
A writer who can't recognize LLM voice isn't a good writer, just like a writer who doesn't do research or fact checking isn't good. It has nothing to do with method, its about skill and work effort.
Ah good ol' serendipity. I strongly believe this too, and I think that's why older (less capable) models fared better: they hadn't fallen into the valley yet. Maybe it's not a valley at all, and is just a cliff after which there is uncanny.
When the earliest models launched I was super-excited to no longer have to write pull request descriptions and commit messages. I have done a sharp 180 on that and I'm no longer a meat proxy[1]: instead I point out any redirects and improvements I made the agent do before opening the PR (which there are still _many_ in even the most trivial tasks).
Given the amount of angst being expressed in this thread about particular LLM-isms, I wonder why people don't simply prompt against the traits they hate. Yes, it will reduce output quality very slightly in some circumstances, but if it really gets your goat like this, surely it's worth it.
Surely someone with a neuro/psych background has picked this up and developed an evidence-backed explanation.
Compared with human writing, all nine frontier and near-frontier models use more inanimate subjects with animate verbs and more low-frequency n-grams, and most also use longer dependencies, more stacked nouns, and more reduced object relatives in conversation. [...] We propose and test a hypothesis as to why: while human language is optimized for successful communication, AI-generated text is increasingly optimized for other goals like math, code, and reasoning, in which there is a pressure for information density rather than successful communication.
AI does the same with responses where the cadence is some type of formula that becomes identifiable.
Each of the big models has their own but an example of the cadence is
A) Agreement or Acknowledgement with the User
B) Specific detail gets a paragraph of detail related to the previous response
C) Overview of the conversation's current state
D) Prompting or Direction for this conversation to go
E) Landing on a beat. Wrap up the vibe. Recap with a zinger.
I cannot unsee it and a few of the models are annoying. I have attempted to build out the Agents and context to avoid the 'last beat one liner' without success because after 2-3 more turns, I get "That challenge was accepted and you stepped up. You're the winner this round".