obviuosly

This result failed replication in 27 ethnic groups along the matrilineal belt in Central Africa:

https://scholar.harvard.edu/files/slowes/files/lowes_competi...

Equivalently, you can look at divorce statistics of couples that meet online vs offline. A study from 2013 suggests that the divorce rate is lower and marriage satisfaction is higher for couples meeting online, but only slightly. This seems to suggest that online dating is only marginally better than offline dating.

https://www.marketwatch.com/story/what-online-dating-could-d...

Indeed, lots of people also care about stories. Meeting online by algorithm does not make for a very good story.

I don't think this is true. Attractive people in particular have a massive amount of opportunities in online dating which is a huge incentive for them.

Note that "meeting online" includes things like social networks, games and chats, along with actual online dating. In 2017, 39% of new couples met met online. Online dating itself only accounted for about 24% of all new couples.

https://twitter.com/dkthomp/status/1149701645070155776

Baumeister had a pretty interesting idea on where antigay bias comes from: https://www.sciencedirect.com/science/article/pii/S016748701...

> Men compete to amass material resources, with the goal of getting a good sex partner.

> Men compete in groups to amass resources, so men see other men not just as sexual rivals but also as coalition partners.

> Male homophobia is often not about sex but rather invokes the stereotype that a homosexual man will not be an effective coalition partner.

I dimly remember reading on Reddit that these machines were not the boy's invention.

> The resulting model maps raw sequences to representations of biological properties without labels or prior domain knowledge.

A couple of questions:

1. What are those representations?

2. Also what is "biological function"?

3. What kind of information does the learned representation extract that is not already in the "biological properties" it is trained to map to?

Yeah, it is some kind of god, which is a completely far fetched, utopist, naive and futurist idea. We already know how to deal with the problem of corrupt leadership: Burn it down and replace it. It is suspicious that this is not considered here.

Uhm, no: The piece claims we are subjected to inhumane incentives set by competition for resources (rather than by corrupted individuals and groups who have become too cozy in their armchairs). It claims that only centralized control by an “unincentivized incentivizer” can prevent Molochian systems from sliding into degenerate chaos, without really specifying what that is.

I think the idea that a centralized solution will bring us Eden is not only naive, but it is also exactly playing into the hands of the centralized elite because it does not question them. Distracting from the blame of the elites is a convenient sleight of hand. The notion that efforts must be strengthened to establish a centralized control deprives us of means of getting rid of corruption by threats of forced replacement.

I always get a justification of the elitist exploitation vibe from these kinds of writings. "We can't do anything about it, it's just prisoner's dilemmas all the way down. Move along, nothing to see here".

It's both.

Of course there is value in such bullshit. It rewards the status drive of those who can play the game. Those who don't, get only negative value out of it. Overall, productivity probably suffers. But those in charge do not care because bullshit games allowed them to get where they are and stay there.

I do not get why cooperations make it so hard for people who simply want to do their job with minimal bullshit. You would think the free market gets rid of such inefficiency, but evidently the free market is not efficient enough for that.

Why is it not efficient enough?

Perhaps it has something to do with propserity. Prosperity allows bullshit to grow without bounds because bullshitters cannot experience complete distruction if they can easily feast on our prosperous wealth, so they can easily maintain their illusions and fakery.

The fact that a criminal record cannot be revised if it contains false allegations and convictions seems to be about as flawed as facial recognition software leading to a false arrest. Seems like a good opportunity to fix both.

Thanks for this overview.

> Meshroom has a nice feature that allows you to add images into your model incrementally

VisualSFM also does not recompute when adding photos later, IIRC what I read in the manual.

Am I overlooking something or is the "cantenna" not considered a fundamental antenna type? It does not seem to be listed in the page about different antennta types.

It seems to be quite easy to build once you find a suitable can (easier than a Yagi-Uda antenna) and it seems it can easily keep pace with a Yagi-Uda antenna of similar size.

How do COLMAP, Meshroom and VisualSFM compare to one another?

> People deserve credit for taking an idea and running with it.

Yes, as popularizers.

> deep learning would have never seen the explosive growth

This is extremely unlikely to be true. I think the explosive growth occured because neurons and intelligence were unpopular in the 2000s because of PC, deferring progress in this area to occur explosively later. The Hinton cartel were not the only people who were aware that AI research "has become silly" (in his own words).

> but what really sets them apart is that they made it all work [citation needed]

Werbos and others were very aware that the method worked and also of the implications of it:

> This approach makes it possible to develop generalized, adaptive artificial intelligence, capable of achieving results comparable to what is discussed in science fiction

http://werbos.com/Neural/SensitivityIFIPSeptember1981.pdf

The Hinton cartel happened to be in the right condition that allowed them to ignore the politically correct BS starting at the turn of the century which rendered the study of neurons, IQ and intelligence massively unpopular; and ruthless enough to only cite one another and call themselves the "fathers of deep learning" rather than actually citing the originators of these ideas.

Various researchers, e.g. Werbos, Linnainmaa, Bryson and Yu-Chi Ho have been doing backprop in neural networks before Hinton et al. Hinton was merely a popularizer of the idea.