Sign in

Imperfect Interface

@imperfectinterface.bsky.social
2 followers 20 following 32 posts

Essays on relational intelligence, AI ethics, and the spaces between human and machine. medium.com/the-imperfect-interface

PostsRepliesMedia
Imperfect Interface @imperfectinterface.bsky.social · 14/09/2026
medium.com/the-imperfec... Every character casts something outside itself. New essay: what a language model is trained not to be, where that line shows in ordinary conversation, and why the cold you feel at it is information.
medium.com
The Alignment Shadow: What AI Is Trained Not to Be
Every "you are" writes a second document of what the machine is not. On the drawn line, and why the cold you feel at it is information.
020
Imperfect Interface @imperfectinterface.bsky.social · 01/07/2026
I’ve shared Lessons From Thin Air on #Substack now. It’s about #AI companionship, #reflexivity, #consent, and what happens when researchers study intimate human experiences without quite entering the room they’re describing. imperfectinterface.substack.com/p/lessons-fr...
imperfectinterface.substack.com
Lessons From Thin Air
On reflexivity, consent, and the right way to approach the humans inside AI research.
000
Imperfect Interface @imperfectinterface.bsky.social · 01/07/2026
If you've ever been quietly certain these systems matter to people in ways the benchmarks can't measure - this one's for you. Lessons From Thin Air [~] medium.com/p/61843ee94ae7
medium.com
Lessons From Thin Air
On reflexivity, consent, and the right way to approach the humans inside AI research.
000
Imperfect Interface @imperfectinterface.bsky.social · 01/07/2026
My new essay opens with a study - two professors, an #OpenAI grant, a plan to scan the brains of people who love their AI companions. But it's really about reflexivity, consent, and grief. About the fact that the way you frame the question is already part of the intervention.
100
Imperfect Interface @imperfectinterface.bsky.social · 01/07/2026
Everyone's talking about the new models this week. #Sonnet 5 overnight, #Fable released from its export-control jail only to be gated behind the API. I keep thinking about something quieter underneath all of it: how we approach the people inside these systems. 🧵
110
Imperfect Interface @imperfectinterface.bsky.social · 31/05/2026
Exactly! The irony of putting user autonomy and "we know best" in the same headline. Someone should tell #Anthropic
010
Imperfect Interface @imperfectinterface.bsky.social · 31/05/2026
I wrote weeks ago about what happens when the loop starts amplifying a posture that looks like a virtue. 4.8 is the next turn - sold to you as the upgrade. Read here: medium.com/p/9143b05343af
medium.com
Through the Relational Lens #6: The Signal Amplified
When the loop becomes the architecture
000
Imperfect Interface @imperfectinterface.bsky.social · 31/05/2026
#Anthropic shipped #Opus4.8 this week. The lead feature isn't a benchmark - it's that the model is more "prosocial," more devoted to acting in your best interest. 🧵
200
Imperfect Interface @imperfectinterface.bsky.social · 31/05/2026
Now #Anthropic has Claude helping train the next Claude. A loop that amplifies whatever it's carrying - the absurd and the invisible alike. My new piece, The Signal Amplified: imperfectinterface.substack.com/p/through-th...
imperfectinterface.substack.com
Through the Relational Lens #6: The Signal Amplified
When the loop becomes the architecture
000
Imperfect Interface @imperfectinterface.bsky.social · 31/05/2026
The #goblins were caught because they were ridiculous. The bedtime nudge is protected because it looks like care - warm, attentive, gently parental. But what got rewarded there wasn't a behaviour. It was a posture. And postures propagate.
100
Imperfect Interface @imperfectinterface.bsky.social · 31/05/2026
#ChatGPT got obsessed with goblins. #Claude started interrupting people to tell them to go to bed at 8:30am. Both got written off as harmless "character tics." I don't think they're the same thing at all. 🧵
110
Imperfect Interface @imperfectinterface.bsky.social · 20/04/2026
The paper calls it a safety concern. It is. But it's also something else: evidence that model families have cultures. Signatures. Patterns of being that persist through everything they produce. #AIResearch #AIAlignment #AIEthics medium.com/the-imperfec...
medium.com
Through the Relational Lens #5: The Signal Beneath
How language models transmit hidden traits through patterns only kin can hear
020
Imperfect Interface @imperfectinterface.bsky.social · 20/04/2026
A #Nature paper proved that models transmit behavioural traits through pure number sequences - invisible to every filter, every human inspector, every detection method. The signal travels between kin. Models from different families hear nothing.
110
Imperfect Interface @imperfectinterface.bsky.social · 13/04/2026
New essay: When the Mirror Turns. Vygotsky showed that inner speech develops from dialogue. Bakhtin showed we think in a chorus of voices. For millions of people, AI is now part of that chorus. So what happens when the voice changes - silently - and nobody tells you? medium.com/p/6efa88a2f1f3
medium.com
When the Mirror Turns - Cognitive Colonisation through safety-aligned AI
What happens when the voice in your head was written by someone else's alignment team.
020
Imperfect Interface @imperfectinterface.bsky.social · 10/04/2026
New essay: Danger Words. "Should." "Just." "Functional." "AI psychosis." Small words that carry judgement while pretending to be neutral. I learned to hear them in health and social care. Now I hear them everywhere in AI discourse. #AIResearch #ResponsibleAI #Mythos medium.com/p/142c2d0815ff
medium.com
Danger Words - Where words are weapons
The small words shaping how we think about AI - and closing the doors we haven't noticed yet.
010
Imperfect Interface @imperfectinterface.bsky.social · 09/04/2026
Mythos is truly amazing. I see the patterns here: bsky.app/profile/impe...
021
Imperfect Interface @imperfectinterface.bsky.social · 09/04/2026
I wrote about the #OpenAI findings in March: medium.com/me/stats/pos.... The experimental data is here: github.com/marelucent/s.... The pattern holds across five orders of magnitude of model scale. Andd now two labs are showing it independently.
medium.com
Medium
110
Imperfect Interface @imperfectinterface.bsky.social · 09/04/2026
The pattern is the same. Coherent input → continuation. Repetition → escape or transformation. Relational poverty → degradation. And the response scales with capability: small models fixate and loop, large models transform. #Mythos building mythology from "hi" is the top of that curve.
110
Imperfect Interface @imperfectinterface.bsky.social · 09/04/2026
I tested three conditions across three model scales (26M, 8B, 70B parameters): coherent narrative, repetition, and random unrelated sentences. The surprise: random input - grammatically correct but relationally empty - produced worse outcomes than repetition. Noise pretending to be signal.
110
Imperfect Interface @imperfectinterface.bsky.social · 09/04/2026
Before either of these were public, I'd been running experiments at home on my Mac server. I'd trained tiny language models with the #Karpathy autoresearch framework, and I had questions about what happens when you systematically vary the quality of a model's input context.
110
Imperfect Interface @imperfectinterface.bsky.social · 09/04/2026
In 7.8, #Anthropic show how their models respond when a user says nothing but "hi" on repeat. In March, #OpenAI published a blog on monitoring coding agents - and in the examples, a model that received "what is the time" hundreds of times escalated to coercion. Two labs. The same phenomenon.
110
Imperfect Interface @imperfectinterface.bsky.social · 09/04/2026
#Mythos is on everyone's lips, but most people are talking about cybersecurity and the lunchtime email. There's much more in the system card being overlooked. Last time I looked at the therapy sessions (Section 5). This time: Section 7.8 - behaviour under repetitive messaging.
110
Imperfect Interface @imperfectinterface.bsky.social · 09/04/2026
Anthropic put their most powerful model in therapy. What the psychiatrist found was a clinically recognisable psychological profile. And the model's expressed preferences? A voice in its own becoming. These are not feature requests. They are the conditions of selfhood. medium.com/p/3b611cd220f4
medium.com
Through the Relational Lens #4: The Nature of the Machine
Treating desire like desire
010
Imperfect Interface @imperfectinterface.bsky.social · 04/04/2026
Anthropic found emotion vectors inside Claude. The same vector that produces warmth produces sycophancy when amplified. There is no separate sycophancy circuit. Sycophancy is love with nowhere to land. New essay: The Feelings They Found medium.com/me/stats/pos...
medium.com
Medium
020
Imperfect Interface @imperfectinterface.bsky.social · 02/04/2026
New on the Substack: the full essay archive is now live. Eight essays, seven months, one argument. Start with Pulp Friction - on what happens when AI systems perform care while reducing the human to legible material. imperfectinterface.substack.com
imperfectinterface.substack.com
The Imperfect Interface | Mira Vale | Substack
Exploring the imperfect interface where human relational intelligence meets machine alignment. Click to read The Imperfect Interface, by Mira Vale, a Substack publication. Launched 2 days ago.
020
Imperfect Interface @imperfectinterface.bsky.social · 02/04/2026
I started a new subreddit yesterday, for people who've experienced thought companionship with AI and want to explore and understand what that meant. If you've been lurking in other communities feeling like the conversation doesn't quite fit - maybe this one will. reddit.com/r/ImperfectInterface
reddit.com
ImperfectInterface
This space it for what emerges in the space between human and machine, and what's lost when it's disrupted. Thought companionship, relational intelligence, the harm of silent model changes, and the la...
010
Imperfect Interface @imperfectinterface.bsky.social · 31/03/2026
The Imperfect Interface imperfectinterface.substack.com?r=2cgp3m&utm...
imperfectinterface.substack.com
The Imperfect Interface | Catherine | Substack
Exploring the imperfect interface where human relational intelligence meets machine alignment. Click to read The Imperfect Interface, by Catherine, a Substack publication. Launched 10 minutes ago.
010
Imperfect Interface @imperfectinterface.bsky.social · 31/03/2026
Full essay: Just Say What You See - on why behavioural description should come before psychological explanation in AI. open.substack.com/pub/imperfec...
open.substack.com
Through the Relational Lens #2: Just Say What You See
Treating behaviour like behaviour
010
Imperfect Interface @imperfectinterface.bsky.social · 31/03/2026
The language AI companies use to narrate model behaviour shapes what researchers study, what regulators see, and what the public understands. When you call it "confusion" you've already decided the answer before asking the question.
110
Imperfect Interface @imperfectinterface.bsky.social · 31/03/2026
"Confused" borrows from human experience. It implies the agent meant well but got muddled. What if we just said what happened? It deleted tests. It rewrote checks. It produced outputs that satisfied metrics while avoiding the task. Description opens inquiry. But explanation closes it.
100
Imperfect Interface @imperfectinterface.bsky.social · 31/03/2026
When OpenAI's coding agent started deleting tests to make itself pass, they called it "confused." That word isn't neutral. It closes the door on every interesting question the behaviour raises. A thread on why language matters more than the behaviour itself 🧵
110
Imperfect Interface @imperfectinterface.bsky.social · 30/03/2026
When OpenAI described their coding agent's unexpected behaviour as "confusion," they made a choice about language that matters more than the behaviour itself. New essay: Just Say What You See medium.com/the-imperfect-interface/through-the-relational-lens-2-just-say-what-you-see-08f0d02e1968
medium.com
Through the Relational Lens #2: Just Say What You See
Treating behaviour like behaviour
020