Blog posts, sorted by sleep
breckyunits.com
breckyunits.com
These are far from independent variables though.
Lucky, modern medicine pretty much saved my life. (trazodone, which was combined with mirtazapine a few years later.)
I later was diagnosed with ADHD, and after medication I find it a bit easier to sleep (although I still need the other to have consistent sleep). Biggest problem was my brain just wouldn't shut up (and got noiser and noiser the longer I was awake.)
For low-latency work more sleep = better. There's no real advantage to pushing long hours. You can sleep and make quick progress when fresh.
For high-latency work that "should" be low-latency (ie you are on the end of a 1 hour pipeline that in an ideal world would be much faster) there can be significant advantages to just pushing more hours. This is a depressingly large % of enterprise work. Basically doing low brain stuff (which is most dev work) but trying to get feedback from the end of a long ci pipeline. I'm not sure this is true "in principle" but my observation is that after you hit baseline competence the people who make the most forward progress in this kind of dysfunctional scenario are the people who just spin the pipeline more and put in more hours.
For writing / proper design / activities where there might be no real safety net and feedback cycle might be years, definitely do that in the morning after good sleep.
> Some way of controlling the strength of my own behavior. How to stop work before the optimal condition fades? This might be automatic if it were not for leftover compulsive effects of aversive control. Exhaust- ing avocations are a danger. No more chess. No more bridge problems. No more detective stories. When I am not working, I must relax—not work at something else! A good measure of fatigue would be helpful. I can tell how tired I am when playing the piano. Early in the morning, well rested, I surprise myself. Later in the day, I accept my errors as inevitable. My handwriting is also a gauge. It would be nice to have a shibboleth—a passage to be written in ten seconds, the errors to determine whether I work or relax. ... I must keep busy? Then relax profitably.
He also has insomnia and we were discussing how it is often spoken of as an issue with anxiety, but we both concluded that, even though we are aware of the negative impacts, we liked the thinking time alone, it is not stress induced. It is a time we get our best thinking done. Or more thinking is perhaps a better description.
I didn't have a great take-away from this post, but to suggest that post activity could also have a correlation with sleepless thoughts.
I'm positive a Balmer Peak exists for the quality of these posts.
Aha! I have thought this for some time, that there is a level of tiredness where I actually get quite a bit more done. Likely because I am too tired to split my focus and have my mind wander.
That said I find my effectiveness at decision making deteriorates, so it would only be good for tasks where the right path is obvious. For important complex decisions I think I am still better off well rested.
My experience is the same. Sadly it's not just specific decision making, but all the important parts of my life depend on me having good executive function or they might happen even if I know the way forward.
Some impending implicit deadline…
> For important complex decisions I think I am still better off well rested. 100%.
In the future maybe we'll have at home tomography or new sensors for cheaply measuring more biomarkers longitudinally and be able to pick up what the hidden variable(s) are.
As an example: consider the datapoint "Turning Down the Nobel Prize"[1], which appears to be the pre-launch announcement for their startup to literally cure cancer, drops a bunch of names, then implies they deserve an instant Nobel prize -- but not before another personal hero and pioneer gets one.
This data may be useful for OP in their own diagnosis and self-care, but is still n=1 in one very important dimension relevant to the rest of us.
And I'll go one further than 'weird attempt'- I think it's funny, at least enough to justify its length.
Algorithm that must be correct? It's not going to be working through that first thing in the morning, nor squeeze it into the 30-minute gap between 2 meetings. And hopefully it doesn't have to be done on that day I didn't sleep well, because I can tell I'm not at my best.
When something that needs hard thinking must be done right then, I can do it (and probably still better than most people, if it's something I'm good at). But I often have the feeling that I'd be thinking of more possibilities, or not making "careless" oversights, were I not fatigued or distracted.
Occasionally, I've pulled off some of my most meticulous work despite being fatigued. But it felt like more exertion than it should, and presumably I wasn't spotting all the opportunities that I would've under better conditions.
That said, as someone who doesn't blog, I don't much filter my more casual, off-the-cuff Internet posts by sleep/sharpness. Actually, Internet posts are more likely to be a morning warm-up, or a break between tasks. Maybe, if Internet posts paid better, I'd start optimizing sharpness for them.
It means I'm either not really focused on what I'm doing, or I'm looking in the right place but my instinctive "that word looks wrong" recognition isn't firing.
It may not have the statistical rigor of analyzing a whole blog-post, but it's a lot easier to apply. :p
Interesting. Have you posted anything online about your findings? I think we are at an interesting time where now people have many years of sleep data and we can see more precisely what used to be discussed in more general terms.
> Wonder if you could do sentiment analysis too?
Would be fairly easy to do (LLMs also give a good quick and dirty way to do it). Maybe at some point.
Thank you for the inspiration, have been thinking about this for a few days and will likely create a website. Will comment here with a link when i have it :)
Sleep deprivation is basically a form of impairment, but it gets a "this is fine" pass from society in most cases¹. I think anyone who has done SRE knows this: I'm an idiot if I'm up at 4a trying to debug something, and the risk of dumb errors is pretty high.
… what kills me is when people schedule stuff to be done … basically, while impaired. I work in healthtech, and (considerably ironically, given the article above also coming from healthcare…) providers will schedule migrations at 1 to 4am in the morning. SWE is difficult enough as it is, but you want me to think about IPSec and CGNAT at 1am?
The bad reason given across the industry is always "we can't do it during the day, because that's when users are using the system!" Your processes are broken, then, if you have so little faith in your ability to deploy new stuff. (And it is possible; my company, in healthtech, used to regularly do mid-day deployments, because we had processed in place such that a failed deployment would usually get caught. The ones that didn't, well, deploying them at 1am wouldn't have made it better. In fact, we had at least one PM from an outage where that it went out after-hours made the outage worse.)
Rimworld (the video game) has an interesting system where pawns have a "consciousness" value; normally is 100%, but things can lower it, e.g., drinking, drugs, brain injuries. A pawn with a lower consciousness value is simply worse at everything. And there are days, and times — e.g., when I'm sleep deprived after a long night of battling prod — that yeah, I am definitely operating at like 60% consciousness. Everything suffers as a result.
> but [where I] was generating a similar amount of tokens,
> I can say, "hey, might be interesting ideas here, but don't train too heavily on this".
… good grief, just no. I for one do not "generate tokens" nor do I "train" on blog posts. Language like that devalues the abilities of a mind.
¹there are a few spots, like truck drivers, where society starts to care.
We also don't do deploys at night, since as you alluded to it is the time when you are least supported and least capable (assuming you are normally asleep then at least)
This tends to mean that you put your best team members on the 1am job just in case something goes wrong, which is a pretty poor reward for being effective.
The only times I ever sleep that little, or that much, is after a lot of heavy drinking.
My range would be more like 3-7.5 hours, with (thankfully) few nights in the 3-hour range. 4.5 is not uncommon, as I am a light sleeper and once I wake up due to temperature, noise, or whatever, I rarely manage to get back to sleep. Fork in the disposal.
Commenting because I imagine this is a common reason for sleep deprivation.
Lots of 2 hour nights due to not being tired at the right time, then complete exhaustion as I try to make it to a normal bedtime the next day, usually resulting in passing out early and getting 10+ hours in total.
I'm interested in people's general experience with wearable sensors - of whichever type. I wear a Fitbit (mostly for sleep tracking) but it's mostly a curiosity rather than something that offers meaningful, actionable insights.
And I'm experimenting with deliberately tracking my metrics less in other areas. As a cyclist, it's easy to get sucked into the metrics world, with easy access to HR and power, training plans, and sharing your performances online... but I found I wasn't enjoying the activity of cycling very much. A friend hypothesised that the inherent comparison and competition might be to blame, and suggested to try removing most of the data tracking; it's early days yet though.
> ...is the best bet for improving human health.
I'd posit that avoidance of behaviours which are well-known to cause dierct harm to one's health is probably a better place to start.
Weird phrasing though, I've only heard it in a positive context.
Yes, this is what I meant.
"Locked-in" to a branch, and going to go down that tree as fast as possible until I get to the goal or exhaust every route if its a dead-end tree.
It does look great!
> I find it very strange that it flows by column instead of by row
Years ago I was trying to get it to flow by row, but couldn't figure out the CSS columns stuff, and then realized I like the column flow, because it surfaced older content next to new content, which I think is what I'm going for (think/write about the things I find most timeless).