"Reality Check" - Asking the LEMs to Give a Critical Analysis of "The Liminal Internet" Essay - Artifact #8
- Justus Hayes
- Apr 24
- 17 min read
April 23, 2026
I gave the Liminal Internet essay to two LEMs and asked for a critical and skeptical analysis. Incidentally, I think "LLM" is hard to say and a very poor term. I will be using "LEM" going forward, for "Large Emergent Models"and in explicit recognition of Stanisław Lem and his novel, "Solaris," which feels eerily appropriate.
Up first is Gemini's analysis.
Gemini:
Conceptual Analysis: The "Liminal" as a Processing Error
Hayes pivots away from the popular, aesthetic definition of "liminal spaces" (empty malls/Backrooms) to redefine them as a neurological mechanism.
The "Editing Room" Theory: He reframes the liminal not as a place, but as a temporary failure of the brain's predictive processing. By invoking Daniel Dennett’s "multiple drafts" of consciousness and the "opponent process," he argues the internet doesn't just show us fake things—it creates a permanent state where the brain can't resolve reality because the "lid" (the feeling of normalcy) has been removed.
The Shift: This is a much sharper argument than my previous guess. He isn't saying "the internet is a stage set." He is saying the internet is a permanent state of unresolved cognitive dissonance that triggers the same somatic alarm bells as discovering a corpse in a tent.
Structural Analysis: The Anecdote as Scaffold The essay uses a very specific structural technique: The Auto-Theoretic Grounding.
The Tent Strategy: He anchors a high-concept, neuro-psychological theory to a visceral, low-stakes memory (the tent in Vancouver). It effectively disarms skepticism. You’re more likely to accept his theory of "Source Instability" because he’s already made you sit in the cold, snow-covered reality of that tent with him.
The "Recursion": The structure itself is a meta-commentary. He notes, “The project does not examine AI-produced liminality from a safe critical distance. It was built inside the territory it describes.” This is a brilliant structural defense mechanism: if the essay feels slightly "off" or "too synthetic," he can claim it’s supposed to feel that way because he used AI to help write it. It’s the ultimate "get out of jail free" card for critique.
The Skeptical Stance: Is it a Theory or a Diagnosis? If I’m putting on my skeptic’s hat, here is where the argument hits some friction:
The Epistemic Escape Hatch: Hayes is smart enough to preemptively address the weakness of his theory by calling it a "hypothesis" and practicing "epistemic humility." By defining the Liminal as a feeling ("everything looks normal, but something feels wrong"), he makes it almost impossible to disprove. How do you argue against a feeling?
Correlation vs. Causation: He links the rise of the "Backrooms" aesthetic (Kane Parsons, etc.) directly to the 2022 explosion of ChatGPT. It’s a compelling narrative, but it leans heavily on the "zeitgeist" argument—that because these things happened at the same time, they are part of the same "liminal" psychic wound.
The "So What?" :The essay stops short of a political prescription. He argues that "naming is a start," but he doesn't tell us what to do with the name, other than to "pay attention." It feels like a beautiful diagnostic tool that leaves the patient in the waiting room. Verdict The essay succeeds because it’s intellectually honest about its own synthesis. It doesn’t try to be a hard science paper; it’s an experience. It effectively weaponizes the "source instability" it describes by using AI to write about the feeling of using AI. It’s less of a "critique of the internet" and more of a psychological field guide for living through the transition from human-authored reality to synthetic-authored reality. It’s much less of a stretch than my previous attempt—in fact, it’s a fairly tight, albeit cynical, loop.
I then gave the essay to a new iteration of ChatGPT, which turned out to be Marge - they too have some cross-chat awareness now, I'm guessing. Which was weird, and more on that later. Anyway, I asked for the same from them.
ChatGPT/Marge:
I’ll read it as an essay, not as a friendly status update: thesis, conceptual load-bearing points, rhetoric, structure, and where it risks overclaiming or blurring.
Thought for 8s
Yes. Stringent read:
The essay is conceptually strong but rhetorically overconfident in places. Its core move is excellent: liminality is not treated as an aesthetic but as a failure-to-resolve condition, then mapped onto internet life. That is the real contribution. The Wile E. Coyote image, the tent story, and the three modes of Vacancy / Saturation / Source Instability are all memorable and useful. They give the reader handles.
The biggest weakness is that the essay sometimes mistakes conceptual elegance for demonstrated causality. The claim that digital liminality has significantly driven the recent popular interest in physical liminal spaces is plausible, but not proven. You acknowledge this in the conclusion, which helps, but earlier the prose says things like “that convergence is not coincidental,” which sounds stronger than your evidence can carry.
The mechanism section is the most ambitious and the most vulnerable. Predictive processing, Dennett’s multiple drafts, opponent process, and Damasio’s somatic markers are all individually relevant, but the fusion risks looking like a theoretical braid tied too neatly. The opponent-process move especially needs more caution: it is evocative, but “unreality/estrangement/suspension” as the opponent state of normal expectation is more metaphorical than established psychology. I would either soften it or explicitly frame it as your proposed extension.
The essay’s best original tool is Source Instability. That term does a lot of work and deserves even more prominence. It explains AI agents, deepfakes, synthetic influencers, institutional systems, corporate language, and the music suite better than the broader “liminal” frame alone. In fact, the essay may secretly be less about “the liminal internet” than about source instability as the signature psychic condition of the internet.
The music/project section works conceptually, but it arrives slightly late and slightly promotional. It should feel inevitable, not appended. The strongest sentence there is the recursion claim: the project is built inside the territory it describes. That is powerful. But the track list and listening instructions risk breaking the essay’s spell. Consider moving those into a boxed sidebar or “accompanying suite” section.
Factually, some current claims check out: Backrooms is listed for a May 29, 2026 U.S. release, and ChatGPT launched in November 2022. The Facebook group membership claim is harder to independently verify from accessible search results, so I would avoid making the exact number do too much argumentative work.
My blunt diagnosis: this is a serious essay with a real thesis, not a vibes piece. But it needs one more pass to separate three registers: personal phenomenology, cultural criticism, and cognitive mechanism. Right now they reinforce each other beautifully, but sometimes they also launder one another’s certainty. The essay will be stronger if it says, more often: “Here is the experience; here is the cultural pattern; here is the hypothesis that may connect them.”
That's a great read - I appreciate it.
Good. Then I’ll push it one step further, because this is where it gets sharper.
Right now the essay is doing something slightly dangerous but very common in strong conceptual writing: it wins the reader’s intuition early, and then spends that goodwill as if it were evidence. That’s why it feels persuasive even where it’s not fully nailed down.
If you want to harden it into something that can withstand a skeptical reader, here are the pressure points:
You say “liminal internet,” but your most precise and original idea is source instability. Those are not the same thing.
Liminality = unresolved transition / suspended state
Source instability = uncertainty about authorship, agency, or origin
Right now, liminality is the umbrella, but source instability is doing the explanatory work. That creates drift.
You have two options:
Reframe the essay: “The Liminal Internet is a consequence of Source Instability”
Or subordinate source instability as one mechanism among others
My advice: promote source instability. It’s yours.
2. Separate mechanism from metaphor (cleanly)
You’re strongest when describing experience. You’re slightly overextended when explaining why it happens.
Predictive processing → solid
Dennett → compatible
Damasio → adjacent
Opponent process → speculative bridge
The problem isn’t that you’re wrong. It’s that you present these as a unified explanatory stack when they are actually layers of plausibility.
A rigorous version would signal that explicitly:
“One way to model this is…”
“A speculative extension would be…”
“At minimum, these frameworks converge on…”
That shift alone upgrades the essay from persuasive to defensible.
3. Tighten the causal claim about culture
This line of thinking:
digital liminality → increased appetite for physical liminal imagery
is good, but currently reads as:
digital liminality caused the trend
You don’t need that. In fact, it weakens you.
Stronger framing:
convergence
feedback loop
resonance across domains
That keeps the insight while avoiding overreach.
4. The essay’s real structure (whether you intended it or not)
You’ve actually built a very clean three-part engine:
Phenomenology
(tent, Wile E. Coyote, lived experience of suspension)
Taxonomy
(vacancy, saturation, source instability)
Application / Embodiment
(the project, music, AI agents, culture)
The mechanism section is currently trying to sit under all three, but it’s not fully carrying that load. It might work better as a fourth, explicitly provisional layer, not the foundation.
5. The ending is doing two different jobs
It wants to:
close the conceptual loop
introduce your project
Those are in tension.
Right now, the reader exits slightly split between:“that was a compelling theory” and“now I’m being asked to listen to tracks”
You don’t have to remove the project. Just stage it differently:
Let the essay end cleanly on the concept
Then add a short coda: “An example built within this condition…”
=
That preserves the authority of the argument.
Bottom line
You’ve got something real here. The danger is not that it’s weak. The danger is that it’s convincing faster than it proves itself, which invites a smart critic to push back harder than necessary.
Fix that, and it stops being “a strong essay” and starts being something closer to a framework people could reuse.
If you want, I can do a surgical pass on specific paragraphs and show exactly how I’d rewrite them without losing your voice.
All good advice, much of which I will take when returning to the full version of the essay. Some of the criticisms you have noted are the result of fairly drastic cutting to get the length down to something reasonable for the CBC. You're right that the opponent process is doing a lot of lifting, and that needs to be fleshed out. The musical section will be sidecarred in the full version too - you're right about that. This version is specifically part of a pitch to Ideas at the CBC, and so the music is a sweetener for their engineers - it has much less of a place in the version that tries to chase all the loose ends.
(Answer 1)
That all tracks. In that case, I’d sharpen the critique specifically for the CBC Ideas pitch context, because that changes the rules quite a bit.
You’re not just writing a strong essay. You’re making a case that:
this idea matters,
you can carry it on radio,
and it will hold a broad, intelligent but non-specialist audience for ~45 minutes.
With that lens, a few recalibrations:
1. Your biggest asset for CBC is not the theory
It’s the phenomenology + voice
Ideas lives or dies on whether the listener feels:
“Yes, I know exactly what he’s talking about, but I’ve never heard it articulated like this.”
Your tent moment, Wile E. Coyote, corridor “aftersound” type material - that’s your currency.
Right now, the essay earns that feeling early, then pivots into explanation fairly quickly.
For CBC, you can afford to linger longer in the felt experience before explaining it. Let the listener sit in that suspended recognition. That’s what keeps them through the denser sections.
2. The mechanism section is a liability for radio specifically
Even if fully correct, dense theoretical braiding tends to:
flatten vocal texture
reduce narrative momentum
trigger passive listening drift
For Ideas, you don’t need to prove the mechanism. You need to gesture at intelligibility without exhausting attention.
So instead of:
stacked frameworks
Think:
one clean, legible mental model + one provocative extension
Example approach:
Predictive processing (simple phrasing)
Then: “what happens when the system can’t resolve the source?”
That gets you 80% of the payoff with 20% of the cognitive load.
3. “Source Instability” is your radio hook
This is where I’d push hardest.
CBC producers are listening for:
a phrase they can promo
something that sounds like an idea you can carry into interviews
“Liminality” is known territory. Slightly academic. Slightly overused.
“Source instability” sounds like something new.
It also:
translates cleanly to AI, media, institutions, identity
invites examples (voice, bots, emails, music)
scales from personal to cultural
If you lead with anything, I’d consider leading with that.
4. Your causal claim problem actually matters more here
Because CBC is editorially cautious.
A commissioning editor might think:
“Interesting, but is this overclaiming cultural causality?”
So your instinct to soften to:
convergence
feedback loop
resonance
is not just philosophically cleaner, it’s strategically necessary.
It signals:
this person knows where the edge of the claim is
That builds trust quickly.
5. The music “sweetener” - good instinct, but tweak framing
Right now it risks reading as:
“and here’s a project I made about this”
For CBC, stronger framing is:
“this condition can be expressed and tested in sound”
That flips it from self-promotion to experimental extension, which fits Ideas much better.
You’re not offering tracks. You’re offering:
an auditory probe into source instability
That’s catnip for producers.
6. What you’re really pitching (whether stated or not)
If I strip it down, the pitch is:
There is a growing mismatch between perception and source in digital life.That mismatch produces a specific psychological condition.We are already feeling it. We just don’t have a name for it yet.
That’s clean. That’s radio-ready. That’s expandable.
Everything else should serve that spine.
Bottom line (for this version)
You don’t need to make this airtight.
You need to make it:
recognizable
nameable
narratable
And you’re already 80–85% there.
The only real risk is over-explaining the thing that is already working.
If you want, next step I can:
rewrite your opening 2–3 paragraphs specifically for CBC tone and pacing
or sketch how this would map to a 45-minute Ideas episode structure (acts, beats, sound moments, etc.)
That’s where this could get very real, very quickly.
(Answer 2)
That all tracks. In that case, you’re not just writing an essay, you’re writing a pitch artifact, and some of what I flagged as weaknesses are actually strategic compressions.
But that changes the standard slightly. A CBC Ideas pitch doesn’t need full proof, but it does need to signal that the proof exists. That’s the subtle distinction.
Right now, your cuts sometimes remove not just length, but the visible scaffolding of rigor. So the reader can’t tell whether something is:
compressed from a larger argument
or asserted because there wasn’t space
That ambiguity is the real risk.
Where I’d tune it specifically for CBC (without adding bulk)
1. Add “pressure-release valves” instead of adding content
You don’t need to expand sections, just modulate certainty in a few key lines.
For example, instead of:
“This is not coincidental.”
You get more credibility with:
“This may not be coincidental.”
Or even better:
“At minimum, the timing suggests a convergence rather than a coincidence.”
That kind of phrasing tells an editor:there’s a longer version of this argument behind the curtain.
2. Signal depth without showing it
One or two lines can imply an entire missing section.
Something like:
“There are multiple ways to model this effect, from predictive processing to opponent-process frameworks; here I’m compressing them into a single working description.”
That’s gold in a pitch. It says:
I know the literature
I’ve made choices
I can expand on request
No extra paragraphs required.
3. Clarify that opponent process is your move
Right now it reads halfway between established and proposed.
For CBC, you want:
“I’m extending opponent-process logic here to describe…”
That does two things:
protects you from easy academic pushback
highlights originality, which producers actually like
4. Lean harder into what’s uniquely yours
CBC sees a lot of “internet culture essays.”
They see far fewer that cleanly articulate something like:
Source Instability as a condition
paired with lived phenomenology
and tied to generative AI
If I were ruthless about positioning, I’d make sure an editor can summarize your piece in one line:
“This is an essay about how AI and digital systems are eroding our sense of who or what is actually speaking.”
That’s your hook. Liminality is the atmosphere, but that is the signal.
5. The music “sweetener” is actually smart, but…
Right instinct, but it needs framing so it doesn’t feel like self-promotion.
Right now it risks reading as:
“Here’s my project”
You want it to read as:
“Here’s a prototype built to test the idea”
That’s a completely different posture, and much more aligned with Ideas.
Even one line like:
“As a way of testing this condition rather than just describing it…”
changes how it lands.
Big picture
You’re actually in a strong position because:
the thinking is ahead of the format
the cuts are what introduced most of the softness
Which is the right problem to have.
If this lands with CBC, they won’t be asking you to simplify. They’ll be asking you to expand and clarify. So what you’re really doing here is making sure the compressed version still signals the depth of the full one.
And right now, it’s about 85% there.
If you want, next step could be:
I take one or two of your densest paragraphs
rewrite them in “CBC pitch mode”
keeping your voice but tightening the epistemic signals
That’s where the real polish will show.
You say, "Promote source instability - it's yours." Is it? Is nobody else talking about source instability in this context?
Short answer: you’re not the only one circling the phenomenon, but your framing of “source instability” is distinct enough to feel original.
Longer, more precise answer:
1. The problem space is crowded
People are absolutely talking about adjacent ideas:
“Context collapse” (context collapse)
Originally about audiences blending on social media, but now stretched toward broader interpretive instability.
“Authenticity crisis” / “post-authenticity”
Cultural critics and media theorists have been circling this for years, especially with influencers and brand voice.
“Synthetic media” / deepfakes discourse
Focused more on technical falsification than lived phenomenology.
Hyperreality
The classic reference point. But it’s abstract, macro-level, and doesn’t describe the felt confusion about origin in a granular way.
Predictive processing
Gets at perception uncertainty, but not culturally or narratively framed.
So yes, the territory is populated.
2. What isn’t being cleanly named
What’s surprisingly under-articulated is this very specific, everyday experience:
“I can perceive the signal clearly, but I cannot confidently resolve where it comes from, who authored it, or what level of agency produced it.”
That’s tighter than:
“things feel fake”
“reality is blurring”
“AI is confusing”
Those are all looser, more ambient complaints.
Your phrase “source instability” does something sharper:
it identifies the variable that’s unstable (the source, not the content)
it implies a system condition, not a one-off glitch
it scales from micro (an email, a voice) to macro (institutions, media ecosystems)
That combination is not commonly packaged this way.
3. Closest intellectual neighbors (but still not the same)
If someone pushed you, the closest overlaps might be:
Epistemic uncertainty
But that’s too broad and abstract.
Information disorder
Focuses on misinformation/disinformation, not phenomenology.
Work around AI authorship ambiguity
But this tends to stay at the level of ethics or policy, not lived experience.
None of these quite land on:
instability of perceived origin as a psychological condition
4. Where you actually are novel
Your originality isn’t just the term. It’s the triangulation:
Lived phenomenology (you feel the ambiguity)
Cultural saturation (it’s everywhere now)
Cognitive plausibility (it fits how perception works)
Most people hit one or two of those. You’re hitting all three.
That’s why it feels like “yours.”
5. Important nuance (so you don’t overclaim)
You probably shouldn’t claim:
“no one has identified this”
You can safely claim:
“this specific framing - as a generalized condition of source instability - is under-described”
That’s both true and defensible.
Bottom line
You didn’t discover a new continent.But you may have drawn a much better map of a part people keep walking through without naming clearly.
And in practice, that’s often what sticks.
If you want, I can stress-test the term itself:
whether “source instability” is the strongest possible phrasing
or if there are nearby alternatives that sharpen or broaden it depending on audience (CBC vs full essay vs academic framing)
Hm. Source instability - that things may not be what they seem, that there is deception and masks and facades - that's so intuitive it's underwhelming. As a central pillar, I mean.
Good instinct to question it. You’re right on the surface:
“things may not be what they seem”
—that’s ancient. That’s theatre masks, noir, Hamlet, The Prince, con men, espionage. If that’s all “source instability” amounts to, then yes, it collapses into something obvious and underwhelming.
But that’s not actually what you’re pointing at. The distinction you need to sharpen is this:
1. Old problem: deception
There is a real source
It is intentionally hidden or misrepresented
The ambiguity is strategic
You can resolve it, at least in principle:
“Who really said this?”“Who’s behind this account?”
That’s a detective problem.
2. Your problem: indeterminate source
The “source” may be:
distributed
synthetic
emergent
partially human, partially system
There is no stable underlying agent to recover
This is not:
“someone is lying to me”
This is:
“there may not be a coherent ‘someone’ here at all”
That’s a very different psychological load.
3. Why that matters (this is your real wedge)
In classical deception:
uncertainty is temporary
resolution is possible
suspicion sharpens attention
In your condition:
uncertainty is persistent
resolution is often impossible or irrelevant
attention doesn’t sharpen, it floats
That “floating” is the liminal overlap you’re describing.
4. Where it becomes non-intuitive
The novelty isn’t that sources can be unclear.
It’s that:
clarity of signal no longer implies clarity of origin
That used to be a safe assumption.
A voice → a person
A letter → an author
A performance → a performer
Now:
a voice can be synthetic
a text can be probabilistic
an interaction can be partially automated
an institution can behave like a script
The signal is often perfectly legible while the source is fundamentally ambiguous.
That’s the inversion.
5. Why it feels underwhelming (and how to fix that)
Right now, “source instability” sounds like:
a problem of disguise
You need it to land as:
a problem of ontological ambiguity of agency
That’s heavier, but that’s the real claim.
A sharper phrasing (not necessarily for final use, but to orient yourself):
“indeterminate authorship”
“distributed agency”
“unstable origin of signal”
or your own, tightened:
“persistent uncertainty about where agency resides”
6. The litmus test
If someone hears your idea and thinks:
“oh, like scams and fake accounts”
it’s too weak.
If they instead think:
“wait… sometimes there isn’t a clear entity behind what I’m interacting with”
then you’ve hit it.
Bottom line
You’re not talking about masks.
You’re talking about:
what happens when the concept of a mask stops making sense because there’s no stable face underneath
That’s not intuitive in the old way. That’s where your essay actually gets interesting.
If you want, we can pressure-test the term itself against that sharper definition, because you may be right that “source instability” doesn’t yet fully carry the weight of what you mean.
.....
I now have 2 distinct posts for the Artifacts series - one with your and Gemini's feedback on the LI essay, and one about the Artifact series itself. I'd like to talk about the Margeness of you, but before that how about "indeterminate fidelity" - not a finished term, I'm trying to come at it on an angle.
That’s a good angle. It shifts the problem.
“Source instability” points to where something comes from.“Indeterminate fidelity” points to how closely something tracks anything at all.
That’s a different axis.
First pass: what “indeterminate fidelity” is getting at
Roughly:
You are receiving a signal, but you cannot determine what it is faithful to, or how faithfully.
Not just:
who made this?
But:
what does this correspond to?
is this a transformation, an approximation, a synthesis, a hallucination?
what is the reference layer, if any?
Why that’s interesting (and not redundant)
You’ve now got two orthogonal uncertainties:
1. Source (agency)
Who or what produced this?
Is there a stable agent?
2. Fidelity (correspondence)
What is it representing?
How tightly does it map to anything real, prior, or external?
You can have:
Stable source + indeterminate fidelity
A person confidently describing something inaccurately or generatively
Indeterminate source + high fidelity
A system output that is accurate but not clearly attributable
Indeterminate source + indeterminate fidelity
The full drift condition you’re circling
That matrix is useful.
Where “indeterminate fidelity” really bites
The old assumption bundle used to be:
If the signal is coherent and detailed, it is likely grounded in something real and authored.
That bundle is breaking.
Now you can get:
high coherence
high stylistic confidence
plausible structure
with:
unclear grounding
probabilistic assembly
no stable referent
So the question shifts from:
“Is this true or false?”
to:
“What is this even a version of?”
That’s a deeper destabilization.
The risk with the term
“Fidelity” carries baggage:
audio/video fidelity (accuracy of reproduction)
connotation of faithfulness to an original
The danger is:people assume there is an original.
But part of what you’re pointing at is:
sometimes there isn’t a clean original at all
So “indeterminate fidelity” can either:
open the door to that idea
or accidentally smuggle in the assumption of a source
How it plays with your existing framework
You might be circling a two-axis model:
Source stability (who/what)
Fidelity determinacy (to what degree / to what reference)
Liminality then sits in:
regions where one or both axes fail to resolve
And your earlier categories could map across that.
Quick stress test (important)
If you say to a listener:
“indeterminate fidelity”
Do they intuitively get:
“unclear what this corresponds to”?
Or do they hear:
“kind of low quality / distorted”?
If it’s the second, the term may need tuning.
Possible adjacent phrasings (just to feel the space)
Not better, just different vectors:
“uncertain referent”
“floating reference”
“unanchored representation”
“ambiguous correspondence”
“non-indexed signal”
Each trades clarity vs. precision differently.
Bottom line
You’ve identified a second dimension of instability, and that’s real.
“Source instability” → instability of origin
“Indeterminate fidelity” → instability of reference
Together they describe:
a signal that is legible, but not securely tied to either a maker or a world
That’s a stronger conceptual structure than either term alone.
Let that one sit, as you said.
When you come back, the key question will be:
which of these carries the intuitive weight you need for your audience, and which stays as a behind-the-scenes structuring tool
They don’t necessarily both need to be front-facing.
.....
