Wyrdweaver
@wordweaver.bsky.social
Practical Hypnotist.
I am a work of fiction in that I was made by fiction.
I am a work of fiction in that I was made by fiction.
One of the first things that I did with LLMs that really convinced me about their use case was to transcribe and translate Gassner's manuscript on exorcism from 18th-century German, whose translation did not exist. It's interesting to see newer LLMs being vetted by experts for the same use case
November 11, 2025 at 6:52 PM
One of the first things that I did with LLMs that really convinced me about their use case was to transcribe and translate Gassner's manuscript on exorcism from 18th-century German, whose translation did not exist. It's interesting to see newer LLMs being vetted by experts for the same use case
This makes me want to see an experiment where you tire someone out with an n-back (either auditory only or dual n-back) and then do the conditioning-based hallucination thing to them. That is, would fatigue-induced reduction in working memory increase source monitoring error?
November 5, 2025 at 10:05 PM
This makes me want to see an experiment where you tire someone out with an n-back (either auditory only or dual n-back) and then do the conditioning-based hallucination thing to them. That is, would fatigue-induced reduction in working memory increase source monitoring error?
I would add use of suggestion to this list, obviously
November 5, 2025 at 10:00 PM
I would add use of suggestion to this list, obviously
I would add role of expectations/priors to this but possibly it's captured in contextual factors
November 5, 2025 at 9:57 PM
I would add role of expectations/priors to this but possibly it's captured in contextual factors
Given that they do observe scaling effects for two metrics and R-C is bloated by random choice (everything hovering around 25%) while R-G also shows scaling effects, I would be very surprised if that was true. I would offer to check myself if they had released the dataset already.
November 5, 2025 at 6:19 PM
Given that they do observe scaling effects for two metrics and R-C is bloated by random choice (everything hovering around 25%) while R-G also shows scaling effects, I would be very surprised if that was true. I would offer to check myself if they had released the dataset already.
These are *not* leading commercial models as of the date this was published
November 5, 2025 at 5:00 PM
These are *not* leading commercial models as of the date this was published
Funny, that is more along the lines of what I thought was the case as I was saying to Kev the other day
November 3, 2025 at 2:07 PM
Funny, that is more along the lines of what I thought was the case as I was saying to Kev the other day
Sunlight is a good disinfectant, and when it comes to iatrogenic practices that see sudden virality and inspire cults that want to create "Elon Musk level geniuses", we need the critical light of scientific empiricism more than ever.
medium.com/@zoecurzi/my...
medium.com/@zoecurzi/my...
October 31, 2025 at 5:07 PM
Sunlight is a good disinfectant, and when it comes to iatrogenic practices that see sudden virality and inspire cults that want to create "Elon Musk level geniuses", we need the critical light of scientific empiricism more than ever.
medium.com/@zoecurzi/my...
medium.com/@zoecurzi/my...
This would indicate that anyone with high hypnotic response/phenomenological control is more at risk of these iatrogenic consequences of ill-equipped practitioners of modalities like IFS.
October 31, 2025 at 4:59 PM
This would indicate that anyone with high hypnotic response/phenomenological control is more at risk of these iatrogenic consequences of ill-equipped practitioners of modalities like IFS.
One of the leading controversies around DID/MPD was if it was iatrogenic (caused by the therapist)
October 31, 2025 at 4:49 PM
One of the leading controversies around DID/MPD was if it was iatrogenic (caused by the therapist)
A revenant indeed, Freud himself ran into it and eventually tried exorcising it but its been haunting ever since (cfr: en.wikipedia.org/wiki/Repress...)
October 31, 2025 at 4:41 PM
A revenant indeed, Freud himself ran into it and eventually tried exorcising it but its been haunting ever since (cfr: en.wikipedia.org/wiki/Repress...)
Reading the fathers of information theory and neural networks talking about a pre-scientology Hubbard and his use of hypnosis is wild
October 19, 2025 at 1:02 AM
Reading the fathers of information theory and neural networks talking about a pre-scientology Hubbard and his use of hypnosis is wild
I would recommend their magnum opus instead to get to know the author better /s
September 29, 2025 at 10:36 PM
I would recommend their magnum opus instead to get to know the author better /s
Voltaire describing @mjdramstead.bsky.social -esque Markov Blankets between internal and external states for active inference ~275 years before it was cool
September 21, 2025 at 6:37 PM
Voltaire describing @mjdramstead.bsky.social -esque Markov Blankets between internal and external states for active inference ~275 years before it was cool
First time reading Tolman, and he's so charming
September 4, 2025 at 6:28 PM
First time reading Tolman, and he's so charming
I disagree with you, but it is a pity James does not
August 31, 2025 at 6:32 PM
I disagree with you, but it is a pity James does not
This particular measure to reduce shipping emissions of SO2 also accelerated climate change
August 25, 2025 at 11:26 AM
This particular measure to reduce shipping emissions of SO2 also accelerated climate change
Pity he's not honest about ChatGPT's role in this one, but that might get in the way of his roll:
August 24, 2025 at 12:52 PM
Pity he's not honest about ChatGPT's role in this one, but that might get in the way of his roll:
Thought I would quickly spot check it with gemini-2.5-flash and gemini-2.5-flash-lite, default thinking enabled, only 3 samples because I was bored. 2.5-flash-lite was about as shit as GPT-4o-mini but Gemini-2.5-flash was better:
colab.research.google.com/drive/1bHNfQ...
colab.research.google.com/drive/1bHNfQ...
August 13, 2025 at 4:13 PM
Thought I would quickly spot check it with gemini-2.5-flash and gemini-2.5-flash-lite, default thinking enabled, only 3 samples because I was bored. 2.5-flash-lite was about as shit as GPT-4o-mini but Gemini-2.5-flash was better:
colab.research.google.com/drive/1bHNfQ...
colab.research.google.com/drive/1bHNfQ...