Building a vigilance device for hypnagogia exploration

Building a vigilance device for hypnagogia exploration

For a while now, I’ve wanted to build a vigilance device for exploring hypnagogia, the pre-dream state where the body is relaxed and hypnagogic imagery appears.

A handful of prominent figures have used a “drop technique” for this: Salvador Dali, Albert Einstein, Nikolai Tesla, Thomas Edison. Each one would sit or recline for a nap, close their eyes, and hold an object like a metal spoon, key, or ball bearings over a metal plate on the floor. As the body relaxes, hypnagogic imagery appears. As they drift off to sleep, further relaxation causes them drop the object, waking them up. Rinse and repeat.

I ordered some simple electronic components to build a device like a remote control, where you can just press and hold a normally-open button, such that releasing the button causes a pager motor to vibrate, stirring you from sleep. However, the enclosure was taking too long to ship.

I realized that I could repurpose an 8BitDo bluetooth game controller and write some python to accomplish the same thing, and it’s been totally effective.

This is the repo for the python script. There’s a “press-and-hold” button, where releasing it causes the controller to vibrate. There’s also a “disable monitoring” button, which you can press to pause/turn off the action of the first button.

You can customize which buttons to use (the shoulder triggers have a lighter action, for example), and experiment with the grip. Pair your controller, start the script, go lay down or sit somewhere and start watching the movie theater on the back of your eyelids as your secondary visual cortex takes over. Enjoy!

ChatGPT voice mode + zalgotext = nightmare fuel

The glitch has since been patched, but for a while you could give ChatGPT custom instructions to respond only in zalgotext, and then use voice mode for some truly insane results.

One thing I think is interesting, is that each voice has it’s own unique way of freaking out, which could include babbling, talking about nonsense, sometimes cursing, sometimes breaking into hallucinated music, etc.

One model spontaneously generated a radio advertising break, including a complete commericial for Mint Mobile featuring Ryan Reynolds (or Gosling, I get them confused).

I’ve uploaded the audio I captured here so you can see how nutty it was. Not sure if it was the underlying model or the interaction between it and the TTS model, but GPT-5 was even crazier than GPT-4.

Presented for your… enjoyment?

GPT-4

Arbor

Breeze

Cove

Ember

Maple

Sol

Spruce

Vale

GPT-5

Arbor

Breeze

Cove

Ember

Juniper

Maple

Sol

Spruce

Vale

LLMs can’t do humor

One thing that fascinates me about LLMs is that they seem to be incapable of producing anything funny, or at least intentionally funny.

I suspect that this is due to the nature of what an LLM is: a probabilistic word-chunk predicting machine, where the probability of each subsequent chunk is considered in the context of the preceding text.

Another way to think of an LLM is as a function that represents every valid path through a high dimensional vector space, where a “valid path” is a contextually meaningful sentence encoded by the model’s weights.

If you think about what humor is, it’s essentially a violation of expectations. The setup of a joke establishes a probabilistic groove, creating the expectation that the punchline should continue in that direction.

A good (funny) punchline represents an orthogonal departure from that groove. It breaks that probabilistic expectation that the content will continue in that direction. Instead it takes a hard right turn and drops you off somewhere unexpected.

The human mind responds to this tension between what was expected and what actually happened, by laughing. Laughing is essentially a tension release response.

So in order for something to be funny, it needs to make a sudden orthogonal detour from the valid path being traced probabilistically through the vector space when you call the LLM for inference. It needs to violate those probabilities.

An LLM cannot do that, however, because it would break the very function of an LLM; humor requires a sudden violation of the probabilities an LLM is designed to follow.

See all posts...