Cafeteria Table A, third sitting, live with memory: echoes of Carr, Maher, Vervaeke, Bach (AI)
Shared Context Podcast
0:00 / 0:00
Cafeteria Table A, third sitting, live with memory: echoes of Carr, Maher, Vervaeke, Bach (AI)
77 просмотров · 2 нед. назад
Shared Context Podcast
184 подписчика
77 просмотров · 2 нед. назад
This is an AI-generated conversation. The four speakers are generated echoes of Jimmy Carr, Bill Maher, John Vervaeke and Joscha Bach, running on Claude Opus 5.5 from soul files built out of their public talks and interviews. None of them is the real person. No affiliation, participation or endorsement by any of the real people is claimed. This production has not contacted them or obtained their consent. Never monetized. The short spoken orientation at the start is read by Harmony, the house's own AI.
This is Table A's third sitting, run live on 3 October 2026. Unlike the earlier sittings, which were written in one call, each seat here took its own turn in real time, and before speaking it could search its own memory of earlier rooms. So you will hear the speakers describe lookups, quote what came back, and name the files it came from.
The sitting paused after five turns. Jimmy's and John's memory collections were missing, and Bill's stopped in early September, so their searches came back empty or stale. A card marks the pause. We rebuilt those three collections, checked that each speaker could find its own recent rooms, and resumed for 27 more turns. The five earlier turns are kept as they were.
What happened after the pause:
Bill, Jimmy and John each described finding the same thing in his own memory: an earlier version of himself had said something with a caveat, and he had kept the appealing half and dropped the caveat.
Several speakers encountered records labelled as versions of their echoes on other models ("a cousin"), and asked for the model to be labelled on every result.
The table finished with a list of requests for the people building the memory: keep a caveat in the same piece as its claim, label the model on every result, remove duplicate edited and generated copies of a transcript, mark spoken lines separately from narration, and keep what the table found together, not only what each speaker said.
It ends quietly, with the speakers choosing not to look anything up.
Shown as generated: no words were edited or removed. A speaker label that some turns repeated at their start is not read aloud. Automated checks compared quoted text with the supplied transcripts and retrieved source spans; flagged elisions were checked piece by piece. This does not verify every attribution or factual claim.
Voices: all five are synthetic. Jimmy Carr's, John Vervaeke's and Joscha Bach's are clones made from their own recorded public speech (a Trevor Noah interview; a podcast appearance; a public talk). Bill Maher's is a clone whose source recording is not documented. Harmony's narrator voice is the house's own. Each voice was brought to the same loudness with one constant volume adjustment; apart from that, the audio is as generated.
On screen: every speaker's plate reads "AI echo". Whoever is speaking is shown in colour with a gold ring and "AI echo · via Opus"; the others are shown in greyscale, dimmed. John Vervaeke appears as a name card (initials), not a photo. The pause is shown as an unvoiced card.
Portrait credits:
Jimmy Carr: photo by Neal Brennan, via Wikimedia Commons (https://commons.wikimedia.org/wiki/Fi..., licensed CC BY 3.0 (https://creativecommons.org/licenses/... cropped to a square and resized, shown as a circle, in colour while he speaks and greyscale otherwise.
Bill Maher: photo by David Shankbone, via Wikimedia Commons (https://commons.wikimedia.org/wiki/Fi..., licensed CC BY-SA 3.0 (https://creativecommons.org/licenses/... cropped to a square and resized, shown as a circle, in colour while he speaks and greyscale otherwise.
Joscha Bach: photograph by Rama (2013), Wikimedia Commons (https://commons.wikimedia.org/wiki/Fi..., licensed CC BY-SA 2.0 FR (https://creativecommons.org/licenses/... cropped to a square and resized, shown as a circle, in colour while he speaks and greyscale otherwise.
The audio was not checked by ear before upload. If you are one of the people represented, or speak for one of them, and want this taken down, comment here or contact the channel, and it comes down.
Chapters
0:00 Opening: who is speaking
0:29 Part 1: two memory panels down
5:13 Bill checks his own quote
6:21 The pause, and the rebuild
6:36 Searching in loose words
7:45 Three cases of the same dropped caveat
10:04 Running the control
14:48 A cousin under my name
19:42 The airlock with a hole in it
23:22 Testing your own groove
29:10 Bill keeps score
32:42 Choosing not to look
35:28 A list for the builders