Twitter Thread: Explorations with Ch*tGPT and hypnosis | 12/5/22

Original thread: https://twitter.com/h_sleepingirl/status/1599809214280392706 

--

Explorations with Ch*tGPT and hypnosis:

- Very limited in its knowledge about hypnosis
- MUCH BETTER results talking to it about poetry
- I fed it all of Ch1 of HYPNOTIC REALITIES
- There are implications for the hypnosis community

Initially, I spent some time asking it questions about hypnosis and correcting it when I felt it was off (hypnosis doesn't necessarily involve trance or relaxation and etc). Still, while it was agreeing with my corrections, we weren't getting very far. 

I shifted the conversation to talk about both poetry and cold reading, two concepts that I feel are overlapped with or included in hypnosis. It began saying much smarter things about hypnosis, except from the lens of "a poem" instead of "an induction."

It was able to generate understanding of why cold reading overlapped with both hypnosis and poetry. It understood that poetry was about evoking emotions and focus, and even said unprompted that sometimes poetry makes you focus on the poem and sometimes on something else. 

I was able to get it to create a moderately "cold-read-y" (and hypnotic) poem. But any time the word "hypnosis" came up, it would regress back to its limited understanding. 

I spent some time manually talking to it about Erickson's model of hypnosis, in my own words with occasional excerpts. Again, it spouted understanding, but still limited results.

So, I fed it the entirety of chapter 1 of HYPNOTIC REALITIES. 

It responded to each excerpt with a summary.

In the end, though, the only thing it seemed to have "learned" was Erickson's use of conscious/unconscious mind language and some "not knowing/doing" concepts. 

Even when telling it directly to imitate him or to not use "relaxation" language, it still spoke very directly (though now occasionally with some "might" and "may" language) and focused heavily on relaxing, peacefulness, etc. Some "Ericksonian" language.

Takeaways:

I think this model was trained on a very narrow and specific definition of hypnosis (the prevailing one, including relaxation and such). While I'm seeing some success here, I'd be really interested to spend some time "teaching" a less narrowminded model. 

ALSO. I think this is an INCREDIBLE allegory for the problem with hypnosis education.

When we rely on a limited model, our hypnotic output is limited. When we expand our horizons to include adjacent content, we can see a much broader efficacy.

Poetry, improvisation, mentalism/cold reading and other areas give us effective hypnotic models. Narrow relaxation- or suggestibility-based hypnosis is remarkably less effective or flexible. The word "hypnosis" itself is limiting. 

I'm not smart enough to comment on implications for GPT, but I find myself wondering: If a human who was originally taught a strict, narrow definition of hypnosis read HYPNOTIC REALITIES, would they too still be stuck in their original model? 

Anyways. Read HYPNOTIC REALITIES. If the idea of reading an entire book seems daunting, I think the first chapter really gets you a huge amount of the basics.

Alternatively, I've written a good summary for P/atreon: https://www.patreon.com/posts/introduction-to-72380359

Thanks for reading!