She had just finished describing to us her (negative) views about James Frey’s bestselling book, “A Million Little Pieces.”

Prof. Harper was the instructor of my Creative Nonfiction class during my undergrad. This class session was about writing dialogue, the imperfection of human memory, and the importance of apparent authenticity when recounting past interactions.

Her main criticism of that book was that the author, by his own later admission, fabricated much of the key plot points of the book, one that he had labeled was a memoir. She presented this in contrast to how one might present a dialogue while recounting an interaction.

She said something to the effect of:

“Don’t deceive your audience. It’s okay if the dialogue isn’t a verbatim recounting, but make a good faith attempt to capture the intent and content of the exchange, and if there is fabrication, acknowledge it as such, for example by saying ‘I remember them saying something like…’”

A Million Little Pieces is, based on the sales alone, a successful book that tells a story. Were it listed as fiction, there would be no issue here – the problem isn’t the content, per se, it’s that the content was presented as if it were true. This is deception, and colors the perception that a reader would have of the experience.

The reader believes that they are connecting with the author through this intimate disclosure of a series of challenging life experiences. The honesty implies vulnerability, that the author is exposing a difficult part of their past. Were it presented as fiction, the notion of “honesty” isn’t a question – by definition it is a series of creative lies, but the reader is allied with the author in this indulgent dream, and the human connection is through the journey the author leads the reader.

Deceiving in Performance

In an Instagram Reel, producer & DJ Deadmau5 describes how the majority of EDM Festival sets, particularly by big DJs including himself, are pre-recorded.

You have to play a prerecorded set. Like EDC I would be surprised if someone played a set that wasn’t prerecorded, to be honest. Because there is such a big deal in the production schedule and the timeline and all that stuff – you know what a dead giveaway to a prerecorded set is? The visuals, believe it or not. It is so immacuately timed that everything is lining up, that means it’s on a playback system. What do they do up there? Twiddle filter knobs and clap? Yes absolutely.

If you search any video site for “fake festival DJ” you will undoubtedly find video clips of many DJs, time after time, standing behind decks, a mixing console, synthesizers, etc. that are either not powered on or are very clearly not actually being used live. The DJs are miming the performance that they might have given were it not pre-recorded.

One of my alter-egos is spinning vinyl. I’ve done it for a long time. I can at least speak with some level of authority here about this. My refrain about DJing is that “I get on stage and play other people’s music for an hour or two, but my artistry is in curating the experience and my craft is in executing it in a way that you don’t realize I’m curating it.”

More concretely: I play an hour or two of records, but I blend them in a way that most people can’t tell where each song begins and ends, so it becomes a gestalt.

I’ve thought about this for a long time and the fact that the set is prerecorded doesn’t really bother me that much; it’s nothing new. Even back in the vinyl days before people had the “Sync Button”, bigger-name sets were prerecorded as a QA measure (and also for syncing stage pyrotechnics / complicated lighting / visuals, etc).

Subverting deception through revelation

The BBC’s show “Top of the Pops” required that bands pretend-perform their songs, which would be played back pre-recorded. The grunge band Nirvana famously lampooned this fact by making it clear they were not playing their instruments. Cobain, the lone guitarist, does not play, or even hold, his guitar throughout the song despite it clearly playing guitar on the backing track.

To that end, Cobain understood the authenticity of human connection, and so the band’s choice to give a superficially inferior performance (he sung his vocals in an intentionally lower register) clearly communicated that this was fiction, allying the audience with the band, and not deception (presenting a false reality under a veneer of authenticity).

Undoing deception through authenticity

What does both me is that the Festival DJs do participate in this veneer of authenticity. It would be more honest for the DJ, or the promoters, to say “DJ So-and-so has pre-recorded a fantastic experience and we have curated top-of-the-line mindblowing visuals to create a next-level audio-visual experience.” The DJ can still be up on stage, throwing cake, hopping around, dancing, T-posing, whatever – they can even still be behind a perfunctory mixing desk if needed, but the pantomiming of controls isn’t needed anymore.

They could lean into this by keeping props (e.g. sheet cakes) back there instead. They could crowdsurf, they could participate more directly with the fans and celebrate “this music kicks ass, doesn’t it? Holy shit those visuals are phenomenal”. They can get on the mic and be their own hype person. It can still be a legitimate performance – in a way it’s even more legitimized by discarding the fiction.

Deceiving with Technology (read: LLMs)

Recently, I used a career coaching service. In the first few interactions I had with the coach, it was very apparent that they were using an LLM to generate the messages. I politely asked them not to do this (between LinkedIn and my normal line of work, I have been seeing so much LLM-generated text I’m finding its “voice” grating).

I had a question for my family and friends.

When does LLM-generated messaging bother you? When are you OK with it?

The anecdotal (N = ~15 or so) responses I informally gathered seemed to coalesce around three qualities.

(1) Generated message from an anonymized / faceless source

The communication is from a faceless organization / team / entity or it was an automated message. People generally didn’t mind that it was generated through LLM generation or other means.

This feels like something we have been collectively conditioned to over years of automated messages. We’ve all received automated notifications from websites, e-mail marketing, SMS marketing, etc. We mentally file it away as spam, and the fabrication history of the message is essentially moot. Personally, I would almost find it more weird if this sort of message was bespoke written for each recipient.

(2) Generated message from a named / personalized source to a broad audience

The communication is from a named person that they didn’t personally know. This was more of an “eyeroll, but I get it” response; there is a tacit understanding that this single person is sending a message to a large audience and that you, the recipient, are one member of that audience.

This one feels like a cousin of the previous point about prerecorded sets for festivals. Realizing that the creation is fabricated feels “less than”, but in a way that is perhaps understandable – we may wish that the content was bespoke created for us, because we can see the point of origin, and perhaps the apparent deception is disappointing, but it at least doesn’t typically feel…hurtful.

Deception requires some manner of a rug pull, and the rug in this metaphor is “human connection”, and the degree to which you are standing on the rug is the amount of intimacy you hold in that particular connection. Finding out that a mass-disseminated piece of content wasn’t actually written by a human is a rug pull where you’re standing on the fringes of the rug.

(3) Generated message from a named source that is personally sent to you

If the communication is from a named person you do know, or if the message was clearly intended only for you, then nearly everyone (I think one person didn’t care) was very annoyed / bothered by the use of an LLM here.

The third point (generated message from a named source personally for you) is the one that where the revulsion was the most pronounced in my (admittedly small) sample audience. It also tracks with the experience I have been having, as well as why I pushed back with the coach in that instance.

N.B. One respondent specifically described that they would be *very* bothered by a text from a significant other or a family member, where there was an expectation of some level of intimacy, but less so if it were from someone where their relationship was purely professional.

Why is it so bothersome?

Deception on social platforms

LinkedIn, which seems to be very polarizing in its very existence, is abundant with posts that are very clearly generated by LLMs. These are often 5-7 paragraphs long. I’ve been seeing similar content on Facebook, as well. In short-form video platforms, there have been more and more videos shared that are very clearly generated (the generating models use very recognizable voices and often the same backing tracks).

Text

The text posts irk me because the person who generated them likely spent less time producing the content than I spent reading it. My kneejerk reaction is “why are you expecting me to give you more labor, to read thoughts that aren’t even fully yours, than you spent creating them?” But even beyond that superficiality, it feels like deception. Frey’s (faux) “memoir” wasn’t necessarily a bad story, but it wasn’t his actual experience. The more painful parts, the parts where the author would presumably be seeking the most empathy from the reader, around institutionalizaion and alleged crimes, were fabricated.

Generated text post on a platform doesn’t necessarily have to be bad information, but it’s not actually your thoughts that you’re sharing. The reader is attempting to empathize and understand your point of view, to sit at the table with you and discuss it, and you’re not there.

Put more practically: I could probably generate the same content by asking a free-tier LLM a similar question, on my own time; the post text is not actually anything new, it’s just fluff. It’s performative. It’s disingenuous. It feels like a one-sided communication.

It also made me consider a future where people are writing text posts using LLMs and everyone else is summarizing them with LLMs, and then humans aren’t even connecting at all. In this case, it feels more like a social betrayal when you step onto that metaphorical rug, expecting to read the real human thoughts of a fellow human, only to later find out that it’s unclear what they think. It’s a betrayal of “I gave you my time and attention, and you wouldn’t preemptively reciprocate.”

Video

The generated videos bother me a little less from a connection perspective, the generated videos feel more like workplace demonstration videos where the creator is an anonymous source, even if they are a paid actor or a real person. It feels more like point 2, above. What bothers me about the generated video content is that I’m just really fucking tired of that stupid AI voice, and also equally tired of hearing segments of the Interstellar soundtrack (a fantastic soundtrack that I am now annoyed by because of its overuse).

Creating video content is very laborious, so, like point 2 above, I understand why someone might use a robot to generate it – but my concerns are then “is this actually accurate? Was this reviewed by any humans? What are the motivations behind the person / people that made this? How trustworthy is it?”

Violations of intimacy

With Frey’s book, by falsely calling it a memoir, the reader is sitting on the side of the table where they think the author is, empathizing with the author’s presented experience, but it turns out that no one is sitting there with them. Were it presented as fiction, the author and reader sit on the same side of the table together, and that’s where the connection is.

LLM-generated text feels like the former – you’re making text and if I engage with it believing it to be genuine, I’m trying to sit next to you at the table, but it’s not actually you there, it’s a robot. When the respondent talked about the difference between a close-relationship and a professional-relationship, this helped me realize that the issue I had there really was about false intimacy, about sitting at an empty table and trying to have a conversation with a note that was left behind. It feels vacuous and isolating.

In my case, with the career coach–a service I was paying for, I might add–I wanted a human connection to help me through a confusing and intimidating experience. I have a Claude subscription, and have already used it for this. I paid for a service because I wanted a connection to another human who can say “Hey man, I’ve been through this, personally. I’ve worked with people who have been through it, personally. This advice is stuff that worked for me, and you’ll be OK.”

Don’t deceive your audience.

…Especially if it’s just one person.