Old man 1: Life is one trouble after another. I'd be better off dead, better yet, I wish I was never born

Old man 2: True, true, but who has such luck ?.. maybe one in a thousand.

My blog:

I'm also building an open source generic ML library: & .... which I guess might be of interest to some people here

Wiki Contributions


Boring machine learning is where it's at

Got Any examples of this being used? I'm always on the lookout for these kind of usecases.

Covid 10/21: Rogan vs. Gupta

Precisely, the mask thing keeps the tourturous forced herding and makes it even worst, it's inhumane.

Covid 10/21: Rogan vs. Gupta

how much protection would allowing children that have at-home parents to stay at home and only come to school optionally for recess to play with other kids give you? A lot more

Covid 10/21: Rogan vs. Gupta

What's the chance that the whole medical research apparatus in the US is corrupt and the vaccine numbers are off? That's the bit that might be underlying Rogan's model which he's not spelling out, that leads to "I don't trust statistics"

Is it 1% ? 10%? 20%? 90%? If so the unkown-unkowns of a vaccine become very high.

Given that it's widely accepted both Russia and China have faked their data, given that he knows people which had issues that were not reported, given previous failures of the health system during the pandemic.

It seems to me that Joe (and most Americans) are irrational for even considering getting vaccinated.

I for one am to deep down into "assume science in the west is generally not corrupt" rabbit hole, if a large % of data in hard science studies were faked my whole world would collapse so I might as well bet on that, most people here are in that camp, but Joe isn't.

Also Joe isn't able to decouple "public policy" from "scientific consensus", neither are most people, even I or most people here can't. Like, I have a few good heuristics like "If it's a simple statistical analysis and the author shared the data with a bunch of people that replicated and n is very high then it's ok" (ala vaccine studies), or "if it's a metaanalysis from cochren then I trust it", but those are less than rational heuristics and I have no idea how an MMA fighter could reach them or why he'd care.

On the whole Joe's response to the pandemic was on-point, I'd be surprised if he'll change his mind past "a lot of people think otherwise so let's meet in the middle". Because that's the rational view based on his (rather dumb and scientifically illiterate) PoV.

Boring machine learning is where it's at

I don't disagree, as I said before, I'm focused on problem type not method.

The fact that human mimicking problems have loads of cheap training data and can lead to interesting architectures is something I didn't think of that makes them more worthwhile.

Boring machine learning is where it's at

Hmh, I didn't want to give the impression I'm discounting particular architectures, I just gave the boosting example to help outline the target class of problems.

Boring machine learning is where it's at

The main issue with surveillance is the same AI vs automation issue.

Once the population is fine with being tracked 24/7 and you have the infrastructure for it:

  1. The need to do so diminishes
  2. You can just make it illegal to not carry a wifi-enabled ID (there's the issue of having someone elses IDs, but so is there the issue of face-covering, and finding incentives to force people into else'scarrying only your own ID is easy)
Boring machine learning is where it's at

I think I was wrong in writing this, and I corrected it on my blog.

What I mean to say was closer to "human mimicking CV" (i.e. classification, segmentation, tracking and other images -> few numbers/concepts tasks). There's certainly a case to be made that image-as-input and/or output techniques as a whole have very large potential, even if not actualized

The LessWrong Team is now Lightcone Infrastructure, come work with us!

Looking forward to seeing projects that come out of this, the LW UX is certainly the most impressive thing about it and one of the few examples of "modern web design" done well that I've seen.

Not being remote seems really weird given the economics of it, but to each their own, I guess.

Contra Paul Christiano on Sex

In sexual reproduction there would be a large pool of correct copies out there and at some point these would be swapped back into this line. With cloning the information is lost for all descendants until random mutation recreates it. 


I think I get your point here, though I think this assumes a lot about how much cross-over mechanisms can actually "detect" genetic damage.

If this damage can mostly be detected only once the organism is mature enough to be selected for/against by "environment" then I think that kind of goes back into the "red queen" style theory that I'm a fan of (i.e. "hidden traits" that occasionally manifest in the population instead of dying out)

If this damage can mostly be detected at cross-over time or when the organism is still very young or in the germ cells themselves... then I'd expect this is also the kind of damage that won't be present in germ cells to being with, or not in many because there's already intra and inter cellular mechanisms to correct for this by inducing apoptosis in the damaged cell.

But maybe I'm missing something and I don't understand the finer details of cross over well enough.

Load More