I enjoyed that a lot, thanks for sharing! Given arbitrary precision, the number of parameters becomes meaningless as a measure of model complexity, and even this one-parameter equation has infinite VC dimension.
Wow, ok, it seems Hacker News rewrote my submission link to use old.reddit.com which doesn’t support the sharing link format given by the Reddit app? I’m sure I submitted https://www.reddit.com/r/midjourney/s/hZgOecxVcK
Edit: yes, exactly, I think what happened is they are replacing "reddit.com" with "old.reddit.com" before following any redirects. On old.reddit.com the link format I submitted is shorthand for "submission", and because the link processor is not logged in, gets redirected to the login page (and finally the URI parameter that would indicate where it came from gets cut off). Ok, next time I will stick to the traditional reddit links here :)
Agreed, their marketing can really sketch one out but CleanMyMac X is tremendously useful. I use almost every function it has once it a while. The app updater for example, it not only updates Sparkle apps but also lists outdated Mac App Store apps. The MAS itself fails to list those updates for some reason most of the time in its "Updates" pane.
These are two awesome posters by Roche showing an overview of important biochemical pathways and processes. They can be explored online and are also available as printed versions (free of charge!).
I agree with your views - its great and ultimately enriching that people approach this fascinating field on these varying levels of mathematical rigor. They complement each other nicely. He kind of says it himself, „we are building tools for people to get things done“, meaning there is this utilitarian point to ML, and empirism as a vehicle into the unknown is totally fine. The rigor/theoretical treatment will come over time anyways.
What I can more sympathize with is this notion of „AI colonialism“ happening somewhat. We can‘t trust our ML tools too easily, and should listen to domain experts maybe a bit more than many currently do.
That he criticizes AlphaFold2 is especially interesting I think. That model is clearly very useful, and crucially it is well calibrated and its limitations are in the process of being characterized. The measures of model quality are also not arbitrary at all, they have been used for sometimes decades in the field. If funding dries up - we will see. That would be tragic as there are still many interesting problems to solve there, AF2 should just be a start that enables research in many interesting new directions.
Another complicating factor is that it is assumed that there are a few different subtypes of depression. For example with and without psychotic features. Another subtypes is the "atypical depression", which actually has, in contrast to classical major depression, "reactive mood" and a high degree of "rejection sensisitivity": https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2990566/
Sure - I should have given some sources in my original comment. The order of my list of purported mechanisms is also not really ordered by significance. I think generally, more and more mechanisms of AD efficacy are discovered, and it can seem that the more we know, the less we know. Additionally of course, there are like 15 different classes of ADs, with hundreds of different compounds in total. What we know for sure I think is that the serotonin and also the monoamine hypotheses of depression are highly oversimplified and even internally contradictory, and many more targets and way more general biochemical mechanisms in the brain must be involved in depression.
- One source covering the 5-HT1A autoreceptor downregulation part in particular there is "Serotonin autoreceptor function and antidepressant drug action" (https://pubmed.ncbi.nlm.nih.gov/10890313/)
- I personally like the books "Antidepressants" by Leonard (that the previous paper is a chapter of) and "Anxiolytics" by Briley and Nutt. These discuss really a wealth of observed/purported mechanisms, also for example including the significance of late gene products.
- Wikipedia: Pharmacology of antidepressants (https://en.wikipedia.org/wiki/Pharmacology_of_antidepressant...) has great sources, especially on the anti-inflammatory and immunomodulative pathways, and introducing HPA axis modulation as another possible pathway.
Yeah, but thats hardly news, we have known that for decades. Otherwise direct serotonin releasers would be the most useful ADs. But these mostly give people nausea and serotonin syndrome. ADs work mostly by being FIASMAs, reducing inflammation and oxidative stress, increasing neurogenesis and neuroplasticity, increasing the activity of other monoaminergic systems, and the serotonergic portion is mostly explained by downregulation of 5-HT1A autoreceptors due to chronicly high extrasynaptic 5-HT levels. This also explains the delayed action of ADs.
At least this is my knowledge from when I last researched this topic in depth.
But that would be a little bit too much for the general public, so it is simplified, to this annoying effect that everyone talks of "low serotonin levels" where in fact the underlying systems are way more complex.
It is unfortunate that NFTs have become synonymous with these (sometimes generative) digital art collection certificates. These are overhyped, useless, speculative rubbish, while NFTs in general are neither a useless concept nor a scam. It simply means that a clearly identified object (could be a real life object, like a bike of a bike rental company) is transacted upon, owned by somebody or otherwise subject to programmatic access by smart contracts. I think NFTs become useful the moment they are connected to the real world. This all applies only if you accept that blockchains have legitimate use cases at all, of course.
I don't feel so at all. The quality of YouTube recommendations (and video essays in particular!) for me at least is higher than ever. Absolutely astonishing at times. I am grateful for all the great content the algorithm helped me to discover, be it in the sciences, music, commentary, anything. Time spent on YouTube actually almost never feels wasted. Also I don't care about the disappearance of the dislike count, I don't need it to gauge the quality of a video. Not to say there aren't any problems with the "algorithm", but overall it works really well for me.