Oh I see. So the hypothesis is "In a healthy animal, stress is a highly-informative signal that inhibits risk-taking. Sleep ensures the stress system continues to inhibit risk taking appropriately."
Makes sense. It's consistent with sleep deprivation raising the level of cortisol and the brain developing a tolerance to high levels of certain hormones.
Oh I think sleep probably plays other roles today! But I don't think those roles require exactly 7 hours of sleep.
And agreed, need to look at long term effects of sleep need reduction too. My vision is more that people have 3-4 nights of 1-2 hours less sleep and then take a break for 3 nights rather than taking a drug to stop sleeping entirely.
We are fundraising for a self-experiment soon!
I think there's a substantial chance that orexin agonists are "just stimulants" and you can't reduce sleep need much with them. But short sleepers prove it's biologically possible and I want to encourage people to start working on this.
The point about children is a good one, I have to think on it more. But it seems consistent with children needing more calories to grow (and they are too young to gather their own calories), so they rest more.
It might be something more complex like: (1) Animals that aren’t careful tend to take a lot of risks that result in them dying. (2) There’s a process that builds up stress that’s about taking less risks. (3) Sleep exists to process that build-up stress and resolve it.
By this do you mean: "when stress builds up, animals take more risks. Risk-taking animals die. Sleep relieves stress, thus enabling less risky behavior?"
It's a reasonable hypothesis, and I'm open to it happening to some degree. But I think all the arguments against "neurons need rest" apply here just as well. Aren't there brain regions more or less exposed to stress? Don't some neurons experience constant (metabolic) stress simply by being active all the time? Are animals with more stress (prey) sleeping more? Do more stressed people adapt by sleeping more?
Why? Orexin-A is perfectly capable of crossing the blood-brain barrier. If you create a gene therapy to produce more of it, it doesn’t need to be produced in the brain.
I should have provided more context, the proposals of Minjune Song and Issak Freeman center on short sleep mutations found in the literature. These are typically mutations of receptors found on neurons in the brain.
Overproducing orexin-A is fine, but a gene therapy for a hormone that can cross the BBB is overkill when you can supply it exogenously right?
By the way, Orexin-A supplementation is the subject of the proposal I mentioned at the end. Should come out today.
Why do you believe that? For Orexin-A? You can buy it these days from a neurotropic store if you want as a nasal spray.
I think you're mostly asking about production costs right? Yes, you can buy orexin peptides and custom RNA, but the cost-per-unit-effect is orders of magnitude higher than small molecules. It's hard to beat the ~$100/kg that you can get with generic drug manufacturing.
Good to have a number for this. Though I think a better counterfactual is between sleeping and actively foraging. Foraging + thermoregulation costs even more calories.
But let's say for the sake of argument that being awake + foraging takes 20% more calories compared to sleeping. Would sleeping actually get selected for? I think so. Evolution can make pretty fine distinctions given enough generations.
For example, cavefish (who live in an environment without light) quickly evolve less pigmentation and underdeveloped eyes to save energy. This is a convergent trait, it's been observed in several different species. Though I'm not sure what kind of energy penalty eyes and pigment have.
1I'm not as familiar with insomnia treatments, but orexin antagonists seem to be an improvement over existing meds. Probably the biggest improvement is the lower risk of abuse and tolerance compared to other medications. Belsomra has been around for over 10 years and seems to be well tolerated and effective. Though it doesn't work for everyone.
The argument that orexin antagonists could help people sleep more without making you sleepy during the day makes sense to me, with one caveat. If the half-life is long enough, the antagonist could block the orexin signal your body normally produces in the morning, making you feel sleepier than you otherwise would have. But perhaps this would be outweighed by the alertness one gets from a good nights sleep.
As for side effects (beyond daytime sleepiness and things we already know to watch for) I guess I would look for changes in motivation. Orexin has a (tenuous) link to reward seeking, addiction, and motivation. If you feel less energy and motivation during the day that might be something to look into. Indeed the point of a good nights sleep is to feel good and energetic the following day, so it's good to check if an insomnia treatment actually delivers on that.
You may also want to consider adding psychological approaches like CBT-i or paradoxical intention. It seems like some insomnia features a self-reinforcing loop of bad sleep leading to frustration which begets more bad sleep.
(ofc, none of this is medical advice)
Wonderful to get more numbers on this!
These examples seem to contradict note 2 where D/N falls for larger C. Now I'm not sure what the trend should be.
It feels like you could derive a rule of thumb based on the loss and the entropy of the dataset e.g. "If my model starts at a loss of 4 bits/token and the asymptote is 2 bits/token, I need X tokens of data to fully specify a model with Y bits stored in the parameters."
Oh that makes sense!
If the predictors can influence the world in addition to making a prediction, they would also have an incentive to change the world in ways that make their predictions more accurate than their opponents right? For example, if everyone else thinks Bob is going to win the presidency, one of the predictors can bribe Bob to drop out and then bet on Alice winning the presidency.
Is there work on this? To be fair, it seems like every AI safety proposal has to deal with something like this.
This is super cool stuff, thank you for posting!
I may have missed this, but do these scoring rules prevent agents from trying to make the environment more un-predictable? In other words, if you're competing against other predictors, it may make sense to influence the world to be more random and harder to understand.
I think this prediction market type issue has been discussed elsewhere but I can't find a name for it.
Yeah I think sleep probably serves other roles, I just don't see why those roles require 7 hours of sleep rather than say 5 hours.
I do agree that basic research is what will actually get sleep need reduction therapies to work at scale. I'm hoping that citizen science and discussion of the topic will encourage more work on this.