I let AI build a tool to help me figure out what was waking me up at night

This reminds me of a weird story...

I went to work at a BBB office once. They turned all their computers off at night and every morning they were back on. It was just "normal" for them.

I can't even remember what problem I was troubleshooting. At the time I was working on IVR systems.

Anwayz, I was working late in their office. Everyone had turned off their computers and went home. At exactly Midnight, every computer in the office turned back on.

I walked around the office looking at desks wondering what had happened. On one persons desk was an alarm clock with a very quiet alarm buzzing. I checked the clock and it was set for midnight (probably a default). About two minutes later it turned off automatically.

I turned off computers and re-set the alarm to go off a few minutes later.

When that alarm clock went off it somehow caused either draw or feedback in the wiring that caused all the computers to turn back on. At the time I wondered if it had something to do with wake on lan.

In any case, I suggested that person take their alarm clock home.

That CO2 concentration looks unhealthy, I wonder to what extent it's affecting your sleep quality (as opposed to waking you up).

> Measure before you fix

In my case, I got a few IKEA CO2 sensors, and after leaving them in the bedrooms for a few days, we found that leaving an outside window slightly open + the bedroom door open, kept the CO2 levels below 600PPM at night.

We're 1000ft/300m away from a motorway, but fortunately the noise pollution isn't bad. So ventilating (even as it's getting cold) turned out to be a simple fix. I hadn't thought of collecting sleep data from our devices, but maybe I'll get an AI to do that, so I can correlate our sleep quality with the environment.

Hey, OP, consider sleeping with ear plugs. They're scientifically proven to reduce night time awakenings due to audio disturbances. [1]

[1] https://academic.oup.com/sleep/advance-article/doi/10.1093/s...

This is really cool. We did a similar thing around 2 years ago but didn't use AI in that case. Just used a phone to record a few nights sleeping. Then a python script. I manually listened for some time in order to find the threshold amplitude (where all sounds would be ignored below and tracked above). Generated a graph that should the spikes of interest. Clicked on the spikes which went to the timestamp in the audio and listened. Not super scientific I know.

Two observations. 1. Often you wake up after a loud noise but like 5 minutes later with no memory of it. 2. even if you don't wake up from the noise your breathing changes, more likely to talk in sleep and shuffle more. So even if you not waking up your quality of sleep is disrupted.

Our case had some random construction like noise in the early morning, lasted around 10 seconds and disappeared. However, we noted even ordinary sounds we didn't think was loud was effecting our sleep.

Solution for that place was earplugs and a loud fan to generate white noise.

I'm surprised that AI didn't tell him that the most likely cause of regularly waking up around 3 am is a cortisol spike. Try some breathing exercises or some other type of stress relief throughout the day, and you might sleep better.

In my case, thinking too much about the causes of bad sleep actually contributed to making sleep worse, so if this guy is anything like me then this whole project could be hurting his sleep rather than helping.

This seems quite over engineered. They could’ve just left their phone recording overnight and done much simpler analysis on the big file. Maybe leverage LLM to write a 20-line python script, at most

> I get the sleep data from my Garmin* watch. Every watch and ring calculates sleep slightly differently, and to be honest, I don't fully trust any of them on the exact sleep stage I was in at any given second.

I love my Garmin, but it's one of the worst smart watches to track sleep with. It consistently ranks poorly in tests that stack it up against pro sleep equipment, and from my experience it struggles to even detect sleep times properly. That 3:32 event that the watch said has pulled you out of deep sleep may not have been real.

My mom would love this one :) .. she told me recently about a long-running chat gpt session that she's had for over a week, where she was going back and forth trying to figure out the source of some strange sound in the building.

I like the temperate graph halfway down the page. It looks like two decaying exponentials alternating every ~40 minutes, with the downward one steeper than the upward one. It's a neat visualization of hysteresis, where the thermostat presumably has a different temperature threshold for turning off or turning on (or perhaps there's a minimum time between state switches). Without the scale it's hard to know for sure.

Related project I did in 2014 tried to do this. I was a web developer so used the web audio APIs to trigger a recording when the decibel level exceeded a certain value. I was living in a big tent in my friends back yard in Sydney at the time and was convinced it was airplanes coming into SYD that were waking me up at 4am but never really captured conclusive evidence because my laptop battery couldn't make it through the night :)

I was under the impression that the pattern "I have a problem -> let's ask AI" is frowned upon here.

Plot twist: the existential dread of an AI-ified world where "AI" is the answer to everything was what was waking him.

Why use a generated image in that weird dirty yellow style when you have a real screenshot to show?

What is the front end built with? It looks nice.

what a waste of technology. you could have had a pen and graph paper hooked up to an microphone 100 years ago and looked for the spikes in the time set.

long time back i had this sense orb that did something similar and it was night sounds made me wake up !

My sleep was not good so I installed panelling and now I sleep better. There you go. Saved you 8 hours and using AI

This is cool, but a simple circular buffer audio recorder connected to stdin would have been sufficient. The recorder records continuously on a circular buffer that stores the last 5 minutes, and whenever OP wakes up, he can press any key on the keyboard to dump the current 5 minutes on storage, with the timestamp as file name. False positives are much less possible, and the whole system can just be a small CLI program.

Earplugs also solve this problem with many fewer tokens.

hint: your watch is probably lying to you and you're following a normal bifurcated sleep pattern.

AI is melting your real world understanding: https://www.sleepfoundation.org/how-sleep-works/biphasic-sle...

This is cool don't get me wrong, but surely overcomplicated? Why not just record audio to disk the whole night then eyeball the waveform for loudness spikes? If you just don't connect it to any network at all, there's no data breach risk (or am I misunderstanding the justification for the noise-detection toggle thing?).

Also the AI-generated hero image looks vile.

We may be entering the age of "disposable software" (some people politely call it "on-demand software"). Until recently, coding was a highly specialised skill and was relatively expensive. So writing custom code for personal whimsy was a luxury only software developers could afford. Not anymore.