Posts

I made an N95-level mask at home, and you can too 2020-11-17T19:00:40.025Z
Book Review: Reinforcement Learning by Sutton and Barto 2020-10-20T19:40:52.736Z

Comments

Comment by billmei on Long covid: probably worth avoiding—some considerations · 2022-01-17T04:59:24.363Z · LW · GW

What does "VOI" mean?

Huge VOI in following behind the crowd

Comment by billmei on Making Vaccine · 2021-02-12T04:25:27.645Z · LW · GW

TIL, thanks for the information on that. I'm not trying to troll, my apologies if my comment comes across that way. It's just interesting to me that this specific scenario was written about before, yet wasn't surfaced in the discussion.

Comment by billmei on Making Vaccine · 2021-02-12T03:46:32.261Z · LW · GW

Is this blog post potentially a source of X-risk? Popularizing the idea of "mix together chemicals you got in the mail" sounds like an attack vector for an AGI wanting to escape its box.

At the risk of arguing from fictional evidence, Eliezer writes in That Alien Message of an AGI coming up with a cover story to convince someone outside the box to construct nanomachines (amino acids!) to do its bidding.

We sent messages [...] to labs that did their equivalent of DNA sequencing and protein synthesis.  We found some unsuspecting schmuck, and gave it a plausible story [...], and told it to mix together some vials it got in the mail.  Protein-equivalents that self-assembled into the first-stage nanomachines, that built the second-stage nanomachines, that built the third-stage nanomachines...

"Make your own COVID-19 vaccine at home!" sounds like a pretty compelling cover story, to the point that nobody on LessWrong has yet commented on the possibility of this being the machinations of a malicious actor, despite this very specific scenario already being wargamed by Eliezer himself!

I'm not accusing the author of secretly being an AI writing this story, I think that's rather unlikely, and a less fictional framing is that maybe normalizing the idea of "it's really easy to try your own biology experiments at home" increases X-risk from accidental or intentional biological hazards created by people who think this blog post is cool and are inspired to try their own experimentation at home.

More generally, if this blog post is an information hazard, what are the norms on LessWrong for discussing or promoting potential attention hazards? Currently we're incentivized to get upvotes for cool, intellectually interesting posts and comments, and we don't get upvotes for keeping information hazards to ourselves. ;)

Comment by billmei on I made an N95-level mask at home, and you can too · 2021-01-24T16:49:28.378Z · LW · GW

They're cut pieces of tissue paper laying on a table.

Comment by billmei on The Best Virtual Worlds for "Hanging Out" · 2021-01-24T16:44:39.789Z · LW · GW

There are more than 50 of these now and I made a list to keep track of them all: https://github.com/billmei/every-proximity-chat-app

Comment by billmei on Covid 12/24: We’re F***ed, It’s Over · 2020-12-31T02:47:19.572Z · LW · GW

Good catch! I watched the section of the YouTube video linked by the citation on Wikipedia, and the original formula they give is this:

We are trying to solve for the selection coefficient, which I interpret as "how much of an advantage does this strain have over the previous strain".

It is here that I realize I don't know how the Wikipedia editor found the 6.4 number, I couldn't find it anywhere in the citation. The calculation they perform with the log odds comes from the YouTube video, which in the cited segment is actually talking about a different lineage, B.1.177 (this is different from B.1.1.7 ! Did the editor confuse these two?)

Reading the slide deck more closely, it says:

Logistic growth model indicates VUI grows +71% (95%CI: 67%-75%) faster per generation (6.5 days)

Limitations: Sample frequency is noisy & overdispersed in ways not captured by this model

So it turns out that this log odds calculation is not relevant to how we get this "70%" number, it was actually simply interpolated from the data by performing a logistic regression.

EDIT: I have now edited Wikipedia to remove the original calculation using the log odds.

Comment by billmei on New SARS-CoV-2 variant · 2020-12-30T20:40:37.264Z · LW · GW

In response to "What does 71% higher growth rate mean?", I made a comment in this thread going through the calculations in more detail.

TL;DR: it appears "up to 70% more infectious" is based on observed data, so that if you previously observed an  of 0.8, you should expect to observe a new  of somewhere between 1.0 and 1.4 with the new strain, ceteris paribus.

Edit: Not sure what is the etiquette around cross-posting comments, mods feel free to delete this duplicated comment if necessary.

Comment by billmei on Covid 12/24: We’re F***ed, It’s Over · 2020-12-30T20:35:52.061Z · LW · GW

In response to "What does 70% more infectious mean?", I found this slide deck, with the relevant part on slide 17, which I will reproduce below (N501Y is the name of the relevant mutation):

For example, under the additive assumption, an area with an  of 0.8 without the new variant would have an  of 1.19 [1.04-1.35] if only N501Y was present

...

For example, under the multiplicative assumption assumption, an area with an  of 0.8 without the new variant would have an  of 1.18 [1.02-1.40] if only N501Y was present

TL;DR: it appears "up to 70% more infectious" is based on observed data, so that if you previously observed an  of 0.8, you should expect to observe a new  of somewhere between 1.0 and 1.4 with the new strain, ceteris paribus.

EDIT: I previously included this bottom section, which MichaelLowe points out doesn't make sense and I also realized is wrong; see comment thread below.

I found the Wikipedia article on the new strain to have the most clear explanation of the calculation, that this is the result of the observed doubling time being reduced from 6.5 days to 6.4 days; I'll quote the relevant bit:

Data seen by NERVTAG included a genomic analysis showed that the relative prevalence odds of this variant doubled every 6.4 days. With a presumed generational interval of 6.5 days, this resulted in a selection coefficient of

They also found a correlation between higher reproduction rate and detection of lineage B.1.1.7. While there may be other explanations, it is likely that this variant is more transmissible; laboratory studies will clarify this.

Comment by billmei on I made an N95-level mask at home, and you can too · 2020-11-18T22:00:05.862Z · LW · GW

Thanks for the reply; I too am not expert enough to quantify precisely how effective surgical masks are compared to N95s, I can only look up research articles which seem to suggest that they are roughly similar, to a first degree approximation. I wrote this post because it seemed like an easy-to-follow DIY guide to making a mask brace is a low hanging fruit idea that has obvious benefits but hasn't been widely disseminated yet. 

I think the keyword is "civilian", in the sense that yes you could technically buy an N95 from the black market after sleuthing around and avoiding counterfeits, but I doubt that this is approachable for the average person. I followed your tip to just google it but every result in the first 2 pages for me was either out of stock or outdated (e.g. a blog from 2018 that hasn't been updated). This may depend on what region you live in, as availability may be more plentiful in some locales, and my SERP is different from yours.

Thanks for the link to examine.com, I like that the article is both comprehensive and nuanced.

Comment by billmei on I made an N95-level mask at home, and you can too · 2020-11-18T16:21:15.465Z · LW · GW

I did the check as recommended in this infographic, which does not involve a smell test. Not sure how I would get the proper materials and setup to do a smell test at home—some smells are gasses which presumably would not be blocked by any mask? You would have to find some kind of material that generates smelly aerosols at precisely the right size.

Comment by billmei on Why don't we tape surgical masks to the face to seal them airtight? · 2020-11-17T19:02:45.917Z · LW · GW

After doing more research into this question, I think the answer is yes, this is actually a very good idea, only it's much more convenient to use a rubber mask brace instead of tape. I summarize my findings as well as present DIY instructions on how to do this at home in a new post: https://www.lesswrong.com/posts/CrikcGiaWK9CSSzrJ/i-made-an-n95-level-mask-at-home-and-you-can-too

Comment by billmei on Book Review: Reinforcement Learning by Sutton and Barto · 2020-10-22T19:16:43.422Z · LW · GW

Yes, this is the idea! My example here is a highly oversimplified description of Rollout Algorithms, a property of Monte Carlo Tree Search, which you can read more about in Chapter 8.10 in the book.