dissonance
room vaasa · watchthe compound eye
the fly
the air
Typed and spoken words become chemistry. Colour = chemistry, grey = human words. Solid = the fly's dictionary (clear), dashed = fuzzy, dotted = lost. A blue dot: it can steer by side. Click a chip for its chain.
the ethogram · last 3 minutes
what the fly asked
the wall
Heard-as are ranks from MusicCoCa against a musical vocabulary, never raw scores. Tag friction in the workshop's Session 2 words.
seams and switches
four latenesses
“Latency can be ignored, tolerated or exploited.” (Chafe)
13:45 go / no-go
what is measured, chosen and invented
A model of a fly's eye and a rule book for its behaviour, not a fly. No body. Not an upload. Today the fly that answers is the rule fly (“textbook”): every rule cites a paper or says it is ours. The fly's eye is flyvis (Lappalainen 2024), a trained model of one eye's centre.
Four ways to reach the fly from /play: rush your hand at the camera (it may jump; from the front row it may back up — a hypothesis), blow on the mic (it may freeze, then groom its antennae — a hypothesis), roll an R or lip-trill (courtship pulse ~35–40 ms apart, a fight ~85 ms), type or say a smell (it goes into the air).
| # | seam | label |
|---|---|---|
| S1 | camera → flyvis (trained eye, model 000) | borrowed (a trained model, Lappalainen 2024) |
| S2 | browser loom detector → escape | invented |
| S3 | turning: from pixels, or from two halves of one right eye used as left and right | invented (ours, not two eyes) |
| S4 | mic → a few numbers (loud, courtship pulse 25–65 ms, fight pulse 60–110 ms, hum, wind) | chosen (a pressure mic is not an antenna) |
| S5 | word → molecule (a curated dictionary + DoOR names; fuzzy = one typo, labelled) | chosen |
| S6 | molecule → receptor (DoOR, n units shown) | measured (DoOR rows) / borrowed (the curated lines: geosmin, CO₂, acids, amines …) |
| S8 | rules → acts (the rule fly) | invented from papers, each cited |
| S10 | acts → prompts (templates; an LLM only if the host turns it on) | invented / chosen |
| S11 | steering sum → the blend dial | invented |
| S13 | fly time vs wall time | measured live |
| S14 | outage → “asleep” | invented, staged — always said as a second line under “simulation stopped” |
| S15 | the fly's sex: male, 25 °C, virgin, one strain (mean life 63.7 days, Zwaan 1992) | chosen |
| S16 | prompts → sound: three generative models trained on other people's music | — |
| S17 | how it was said (only when the host turns prosody on): each spoken phrase → pitch range, pitch against your own last phrases, rising or falling end, syllable rate, pauses, voicing, loudness, breathiness → words like hushed, urgent, laughing, sung, rising, halting → the scribe's prompts and a vocal drive for the fly's arousal/dopa | invented (plain thresholds, ours; not a trained classifier) |
| S18 | the room's own sound as one of the bed's two corners (only when the host turns room_corner on): every few minutes one seated laptop's mic, 10 seconds, rotating seat by seat; its ear button glows while it listens, and the seat's facets light on every screen | chosen |
Moods (hunger, aggression, arousal, dopa, sero) are named analogies, ours: a real fly's dopamine and serotonin are not simulated. The heart tide (forward ~14 s, back ~5 s) is after Wasserthal 2007, numbers via the Inside JEB summary (only the direction is from the primary abstract); ageing of its rate is from Klassen 2017's flies at 20 °C, read at the same fraction of mean life as our 25 °C fly (ours); combining them is ours. A real fly takes off ~215 ms after a loom; this one is slower, and the clocks above say by how much.
Your camera sends a 16 × 16 grey picture 5 times a second; it appears in the fly's eye and here, visible to anyone watching online, and is never stored. Your mic sends a few numbers 10 times a second, never sound, with one exception the host can switch on (S18): then, every few minutes, one laptop in turn sends 10 seconds of the room as it hears it. The door keeps it in memory only, never on disk, and only while it is one of the bed's two corners (a minute or so, until the next clip takes the corner); only the music machine can fetch it, and that machine writes it to one temporary file that the next corner on the same side overwrites; your ear button stops it at any time. When the host turns prosody on (S17), each phrase you speak also sends about a dozen numbers about how it sounded, never the sound. What you say becomes words on your laptop. Your words go to the fly's air and the wall with the name you typed; if the host turns the language-model paths on, they also travel without your name through AHO's course gateway, which keeps a log of every request (how long is not stated: unknown).