3,898 karma · joined February 20, 2012
You can also see some of the pastels I drew and animated here: http://fooladder.com
Email me at yankov.alex at the popular electronic mail service from Google.
My friend has been tracking them since September 2023 here: https://trackingai.org/ . GPT4 seems pretty stable over time, but Llama-2 for example got more conservative in November 2023, and stayed there with only a brief reversion in February 2024.
> Nights were for other worthless extracurriculars to pad out our applications. ... The worst part was knowing that it was all going to be extruded into a few lines in an application form, that a committee would review for about ninety seconds
This is the tail wagging the dog; work really hard for 4 years in HS so that the next 4 years will be spent at a comparatively more prestigious college. I don't think we should expect high school students to be able to optimize this correctly themselves, and from my experience guidance counselors weren't particularly helpful. College admissions feel like a local maxima that have a lot of unintended side effects but would be difficult to change.
Personally, I didn't do the extracurriculars that would look good for college, but rather the ones I enjoyed; I didn't study very hard either. Didn't get into Stanford or an Ivy, but I look upon my experience in high school/college fondly rather than bitterly. Life seems to have turned out ok too.
Location: New York, NY, USA
Remote: Yes
Willing to relocate: No
Technologies: python, javascript, typescript, java, ruby, django, flask, celery, react, postgres, spanner, kubernetes, AWS, GCP
Résumé/CV: https://www.dropbox.com/scl/fi/dcyfj8y5w15qk2jijwenb/Alexander-Yankov-Resume-2023-07-03.pdf?rlkey=hzado5uym118mfi0orf01slux&dl=0
Email: contact info in resume
This disintermediation was a worry, but I don't think this happened frequently enough the justify the fear. It was hard to discern how much this was happening from the data; anecdotally it was rare.
I was working at its competitor Handy then and it was clear that the dealsite method of customer acquisition was non sustainable. At the time Groupon was super popular. It worked like this: a service provider eg Homejoy would sell a coupon through Groupon for $50 for a "regular" $100 service, $25 would go to Groupon and $25 to Homejoy, Homejoy would also pay $45 to $60 to the cleaner. This made sense for some business like a spa that after getting the customer in the door might upsell a $100 massage (and get 100% of that). It did not work with Homejoy because customers who got a deep discount on a cleaning did not retain and did not get upsold to. Since Homejoy had to pay the cleaner, they would loose $20+ on each such "acquired" customer. LTV was negative.
On top of this, Homejoy needed to raise funds and for that you need to show customer growth. If you spent $1000 to acquire 50 customers one month and none of them retained, you need to spend $1200 to acquire 60 customers the next month, or you won't have growth to show investors. Homejoy needed to spend more and more each month or it wouldn't get its next funding round; which it didn't. I don't know but I believe that Homejoy collapsed at its largest size with the peak number of monthly customers.
Other cited reasons for Homejoy's failure like worker classification lawsuits and worker retention are unlikely to have been the real reasons behind its collapse. Handy would have had the same issues and yet they are still going 7 years later as part of Angie's List with little noise.
[0] https://www.forbes.com/sites/ellenhuet/2015/07/23/what-reall...
Currently, and for quite some time now, Horn of the Abyss mod has been the premier way to play single player and the most popular way to play multiplayer HoMM3. Really excited to see how the mod (HotA) and the engine remake (VCMI) projects will interact. Will they find a way to work together or will they split the community?
I looked briefly through rather lengthy minutes [1] of the meeting. Objections seem to come down to usage of words like "bitch" and phrases like "god damn" that students would get in trouble for using in the hallways according to existing school rules. The participants seemed quite queazy about saying the objectionable words themselves and instead either referred by their abbreviations or spelling them out by letter.
One part of the discussion that was also interesting was they considered censoring the book with whiteout for the offending words and one naked drawing but decided against it. They thought they need to get permission from the author because of copyright laws. I believe these kind of redactions are considered fair use for education under copyright laws.
[1] https://core-docs.s3.amazonaws.com/documents/asset/uploaded_...
For password use case, it seems nice that you don't have to store client secrets (eg encrypted salted passwords) on your own infra. However now instead of authentication happening between your own servers and the users browser, there is an additional hop to the SaaS and now you need to learn about JWT etc. At my previous company, moving a Django monolith to do authentication via auth0 was a multi month project and a multi thousand line increase in code/complexity. And we weren't storing passwords to begin with because we were using onetime login emails links.
Maybe SaaS platforms are worth it for social login? I haven't tried that, but I am not convinced that auth0 or some one else can help me connect with facebook/twitter/google better than a library can.
The other day I was casually reading the wikipedia article for Commodus[1] the baddy Emperor from the movie Gladiator and noted how peaceful his rule was. His only wars were when he was co-emperor with his father Marcus Aurelius, and ended when Commodus became solo ruler. By contrast, Marcus Aurelius[2] entire reign was at war and and he is marked by history as the last of the Five Good Emperors.
Wonder how much we think of Roman Emperors comes from the writing and propaganda of their contemporary elites. Rome had an economy based on the back of slaves [3] captured in war after all, so it would not surprise me that elites would praise hawkish Emperors and deride the doves.
[0] https://en.wikipedia.org/wiki/Nero#Military_conflicts
[1] https://en.wikipedia.org/wiki/Commodus#Solo_reign_(180%E2%80...
[2] https://en.wikipedia.org/wiki/Marcus_Aurelius#War_with_Parth...
[3] https://en.wikipedia.org/wiki/Slavery_in_ancient_Rome#Trade_...
I have been running a book club for the past year and a half and that is a fair summary of how it goes. It is definitely hard to keep a discussion on topic for more than a couple of hours without it becoming a generic social event, especially if there are late arrivals. Productivity of work meetings tends to approach zero after an hour; so why should it be different for book clubs? Denser books should require more discussion, but doing it in a single session is hard. The social aspect is fun and is a part of why I run the book club I do, but it is a competing priority with the book itself and different people in the club prioritize the two differently. The prioritization changes for everyone as time progresses and the amount of alcohol consumed increases.
Recommendation of having smaller meetings while the book is in progress deserves a trial. I will adopt it for the next book in my club. While we are holding these virtually, it seems that people are having easier time parceling an hour of their evening time so this will be practical.
There are a few things that differentiates my book club from the authors description of how a typical book club operates. Firstly, I select the book that I want to read and I think my friends will enjoy that is of appropriate time investment. This means there is no griping about book selection and having to competing desires of multiple people. Secondly there is a set schedule: 7:30pm on the first Monday of every other month at my apartment. This means there is no griping about choice of venue (an issue in NYC) and rescheduling to fit prior commitments the participants have. I make an announcement on Facebook and people have a choice to participate or not. This has been working out rather well with turns outs of somewhere between half a dozen to a dozen friends depending on the book. Conversations have been good and organizational investment very reasonable.
[1] https://reframingrussia.com/2020/04/06/covid-19-disinformati...
[2] https://www.research.manchester.ac.uk/portal/en/projects/ref...
With that said, I highly recommend reading Master and Margarita as it is my favorite novel. It is a work however that is deeply rooted in its time and place, so looking at annotations[1] that explain the subtlety is very useful.
This comment was based on my previous two comments here[2][3].
[1] http://www.masterandmargarita.eu/en/
Regarding your search for a non-video link in the grandchild comment, you are in luck. Read about seximal here [2] and make sure to click through the hamburger menu, that's where the meat of the website is at. Unfortunately the website and the video content is not fully overlapping so you kinda have to watch and read both for the fullest understanding of seximal.
[0] https://news.ycombinator.com/item?id=10285392
As for Skripal, he allegedly blew the cover of 300 Russian agents. That is a lot of people with a personal motive. Perhaps an investigation is in order before Britain and Russia tear their relationship apart over this.
5 million figure for Russia for example appears on wikipedia [1] where it is cited from IB Times[2]. IB Times[2] cites Homeless World Cup[3] as the source. Homeless World Cup[3] cites the original IB Times[2] article as the source for the 5 million number! This was the first number I investigated and it already seems super suspect.
[1] https://en.wikipedia.org/wiki/List_of_countries_by_homeless_... [2] http://www.ibtimes.co.uk/homeless-day-2014-new-york-manila-a... [3] https://homelessworldcup.org/homelessness-statistics/
True in my experience when I asked for my comment to be deleted.
> Complaining about being downvoted is discouraged and usually results in even more downvotes.
It works sometimes however. My comment [1] was controversial and was moving up and down between negative and +4 votes for the first two hours of its life. After I made the edit complaining about the downvotes, it started rising rapidly and is now resting at +22.
> Hacker News encourages a single discussion on a given story. All others are marked as a [dupe] and will be killed without the ability to vouch.
An addendum is that from my experience, it is unclear which story the mods will consider the dupe, sometimes the earlier posted story will be called the dupe.
> Relatedly, moderators can also invite users via email to resubmit a post which didn't get much traction.
This happened to my "All the goodness gone from tea (1688)" [2] submission. I am amazed that a moderator read a submission that was only +2 in its original form and also had difficult ergonomics because it was on google books.
> One popular "trick" for obfuscating voting manipulation ... This trick doesn't actually work.
So, undetectable vote manipulation hasn't been detected and there fore doesn't exist? I have trouble believing this. This seems more like a call to not even try claiming that it won't work, because it just might work. I wish some charitable soul with gray morals would step up with their story of voter manipulation that wasn't detected. Tooling could be developed to curb its use by other individuals. They can do this privately to the mods.
Everything is linear if plotted log-log with a fat magic marker.
[0] https://en.wikipedia.org/wiki/Expansion_timeline_of_the_Mosc...
[1] https://en.wikipedia.org/wiki/Proposed_expansion_of_the_New_...
You are being hyperbolic here, but cones around a puddle warrants a discussion.
There have been cases where I have seen portions of a station cordoned off with tape and perhaps with safety cones; however it is not my experience with NYC subway puddles in general. It feels like almost half the stations have leaky ceilings that you can see dripping during rainy weather. No safety cones. During drier days you can find these spots by stalactites and straws forming on the ceiling and by discoloration and mineral deposits on the platform. Two nights ago I noticed something much more frighting on the south bound Prince Street platform. Drip from the ceiling was freezing on the platform and formed an ice-patch a meter in diameter. There were no cones and the subway platform is not the place where I would want to slip. This is a safety issue especially as the whole platform feels only 3 meters wide.
You and I may see MTA workers milling about and appearing lazy to us, but the example of putting safety cones around a puddle doesn't pass the sniff test to me.
> Mr Donald's telegram is from 5 June, and he says his source was someone who "was passing on information given him by a close friend who is currently a member of the State Council".
Here we have a telegram by a guy (British Ambassador) who heard from a guy (unknown) who heard from a guy (unknown State Council) facts about the events of the day prior (massacre was on June 4th). Where did the unknown State Council official get his estimates from; were those official or just something he heard and repeated (and when did he get them)? Initial estimates of disasters are often quite wrong; here they were produced in game of telephone in a day or less; and they are not collaborated by any evidence we have now.
I rank the quality of new evidence as low. Rumors repeated in old official telegrams are still rumors. I expected BBC to have reported more critically. Alan Donald is still alive; BBC could have asked him if he received any updates to that first number that he trusted more.
I also have to fault BBC for it's phrasing around Donald's source. At first reading it sounded like Donald's source is an unnamed member of the State Council who is a close friend of the Ambassador. After reading BBC's sentence a carefully however; it sounds like the Donald's source is a person who is a friend of an unnamed member of the State Council. This ambiguous sentence is deceptive.
EDIT: I see vote count moving up and down on this comment making me think it is controversial. If you disagree with my doubts on the veracity of this story, write a comment. Maybe I missed something.
> Even moderate deficiency, especially in pregnant women and infants, lowers intelligence by 10 to 15 I.Q. points, shaving incalculable potential off a nation’s development.
> All of them — Kazakhstan, Turkmenistan, Tajikistan, Uzbekistan, Kyrghzstan — saw their economies break down with the collapse of the Soviet Union.
Yes, Kazakhstan is a developing country according to the IMF and those who have seen Borat. Using it as an example of the public health success going on in Central Asia is disingenuous however because Kazakhstan is remarkably wealthy compared to its neighbors[0]. It's GDP PPP per capita at $25,263 is only a little behind the digital darling Estonia's at $29,364. Kazakhstan's neighbors like Kyrgyzstan and Tajikistan have theirs at $3,551 and $2,980 respectively.
A lay western reader not knowing is particularities of Central Asia could easily make assumptions and walk away from this article with erroneous ideas. The article, in my reading and without the context, suggests a iodine->health->wealth causality. In Kazakhstan's case oil->wealth->health would be more apt.
[0] https://www.google.com/publicdata/explore?ds=d5bncppjof8f9_&...
I don't know why QZ thought it is a good idea to start a "Meet the man" article with a large photo of a man who is not the titular man.
[1] https://www.citylab.com/environment/2017/09/lab-report-shari...
Its author, Maxim Lott, is a journalist and has been tweaking the calculations over the past couple of weeks reflecting the latest changes in the bills. You may have seen his predictive markets analysis site [1] in the run up to 2016 US Presidential Election.
As a matter of disclosure, the author is a friend and I helped him debug a couple of small issues.
> To take effect, the agreement on a high-latitude fishing ban must be signed by all the countries involved: the United States, Norway, Denmark, Canada and Russia, which all have Arctic shorelines; and South Korea, China, Japan, the European Union and Iceland, which operate ocean trawling fleets.
Thank you NY Times for actually listing the countries involved so succinctly. Unfortunately I have to fault your reporting too. Treaties don't take effect domestically until they are ratified by the involved countries legislatures, not are merely signed. Additionally some treaties have clauses so that they don't take effect until the treaty is ratified by some subset of signatories or by all signatories. It isn't clear which is the case here, but the treaty certainly won't take effect from merely being signed by the nine countries and the EU. It may take years to get the various countries legislatures to ratify. US for example has signed but hasn't ratified the Forced Labour Convention of 1930 [2] or a more recent Convention on the Rights of the Child of 1989 [3].
[1] https://www.nytimes.com/2017/11/30/world/europe/russia-arcti...
[2] https://en.wikipedia.org/wiki/Forced_Labour_Convention
[3] https://en.wikipedia.org/wiki/Convention_on_the_Rights_of_th...