jrfinkel

(Author here.) I completely agree on the cognitive dissonance of whether lockdown was the right move. I don’t even think it’s just politicians, but also people. Everyone has invested so much in it, that the idea that it actually caused net harm may be too much for people to bear. Norway’s health chief admitted it was a mistake, but is the only example of that I’ve seen so far.

I also agree that it jives with Levitt’s work, as well as Karl Friston, and I like they kind of all get to the same place via different techniques.

The most prescient, and overlooked researcher on this currently is Sunetra Gupta of Oxford. This interview with her is over a month old, and she talks about a lot of this stuff: https://unherd.com/2020/05/oxford-doubles-down-sunetra-gupta...

You must not be trying very hard. Open the report and zoom in on the graph if you want better resolution. Its pretty clear that the red bars are deaths, the blue bars are cases, and the grey curve is the R0 (range).

How is it anti-science when they are literally quoting a report from the Norwegian public health authority: ‘It looks as if the effective reproduction rate had already dropped to around 1.1 when the most comprehensive measures were implemented on 12 March, and that there would not be much to push it down below 1… We have seen in retrospect that the infection was on its way down.’

Are you saying that their public health authority is anti-science?

I don’t think anyone is claiming that it suppressed and contained itself. They’re claiming it’s more widespread and more mild. There’s a good graphic in the economist article showing two different peaks that look the same early on.

One of the study authors addressed this on Twitter. It’s certainly not hard proof, but 8 million extra cases would be a whole lot of hypochondriacs. Thread is short and worth a read. https://mobile.twitter.com/inschool4life/status/124762000834...

I don’t think it’s that simple about the tests. They’re less accurate than people think. Here’s one article about it (if you google you can find more): https://www.wsj.com/articles/questions-about-accuracy-of-cor...

Also, if you don’t trust the WSJ either (I wouldn’t blame you) here is an article from Nature Medicine that has a comparison of nasal vs stool tests for covid, and Figure 1b really drives home the high rate of false negatives for nasal swab tests compared to stool tests (and throat swabs are even worse than nasal swabs): https://www.nature.com/articles/s41591-020-0817-4?fbclid=IwA...

Here’s a preprint of the actual paper: https://www.medrxiv.org/content/10.1101/2020.04.01.20050542v...

"It's easier to ask forgiveness than it is to get permission" is actually a Grace Hopper quote, so not Silicon Valley. Depending on the circumstances, it can be good advice.

And "move fast and break things" was about code, not about the world at large. It's a coding "philosophy" for getting shit done, that worked very well for Facebook and did not actually harm people.

"Some will abuse this." LOL, yeah I'm gonna deal with 9 months of pregnancy, and then several months of recovery, and then raise a kid for 18+ years just so I can get that sweet, sweet leave.

Do people not realize that leave benefits are necessary if you wish to attract high skill workers?

Glad you're liking it! I'm one of the backend engineers who designed the system. We devoted a good chunk of time this year to improving our relevance algorithms, and I actually gave a talk on exactly this at Strangeloop in September. The video is at http://www.infoq.com/presentations/machine-learning.

If you want an invite, this link will work for about 16 more hours: http://getprismatic.com/inviteportal?ref=jrfinkel

Merging is pretty simple, and could probably use a little more TLC. The way we do it is that when we get a new document in the system, if the similarity score is above some threshold for documents in two different clusters we will consider merging those clusters. We then make the yes/no decision by comparing random documents from both clusters and averaging the scores, but the threshold we use here is a bit lower than the non-merging decisions (since we have the additional information of this new document doing a good job of linking the clusters).

1. We do ingest multiple documents in parallel, but we also have a system in place for merging clusters, so it ends up not being a problem. 2. Pure hackery. Tweak, look at data, repeat.

Sometimes they're just rewrites of a press release - and those ones are easy to get right - but a lot of the time they really are totally different articles about the same event. Go back and look at the three sets of example clusters and you'll see what I mean.

It's an in-house system similar to the one i built in grad school (http://nlp.stanford.edu/ner/index.shtml).

We actually do weight them differently, but the post was already pretty long so I had to gloss over some of the details.