Having been on the HN frontpage several times with my small static site, it doesn't even show up in my monthly billing reports. The main fluctuation is for when I renew domains.
12,382 karma · joined May 31, 2012
I didn't make this, I saw it on Twitter and posted it.
I think this is what it does:
- gets list of stargazers (starring users) from Github
- checks user's profile for Location
- uses geolocation via Mapbox
- renders map with Mapbox
A lot of this happens in your browser, so if the repo has many stars they may load progressively. On the other hand I haven't seen it error out.
The Mapbox geolocation API has some quirks - it seems it always returns coordinates for a location, even if it's a random string, so you get things like "somewhere else before" being in the Phillipines, or "Okinawa, Japan" being in Florida. (I guess that has to be some weird US base thing, but still...)
There is definitely some weird JS on the input box, but I am able to type "-" without issue.
茶の湯とは ただ湯をわかし茶をたてて のむばかりなる事と知るべし
Tea is naught but this; First you heat the water, Then you make the tea. Then you drink it properly. That is all you need to know.
https://crd.ndl.go.jp/reference/modules/d3ndlcrdentry/index....
Kanji are often made of up other kanji stuck together. In most electric dictionaries you can search by radicals, but not all kanji are radicals; radicals are typically only minimal, indivisible units. This means that even if you recognize the left and right halves of kanji, for example, you may not be able to search them in a conventional dictionary. Examples of characters where this could be an issue are 蹄, 榎, 鮗.
Kanji Club works around this using kanji composition data, so kanji can be made up of arbitrary other kanji rather than just radicals. As a technical note, the preprocessed data is still only around 2MB, so all the search is in the browser - the whole site is static.
https://commons.wikimedia.org/wiki/Commons:Chinese_character...
I've already found it pretty useful as is, though there's still much left to do. Would love to have any comments on it.
https://web.archive.org/web/20190527041657/http://people.csa...
https://news.ycombinator.com/item?id=20019355
This Twitter thread also has a lot of good stuff.
https://twitter.com/maite_taboada/status/1086415051127308288
> Sometimes the results produced by GPT-2 and its ilk are quite startling in their apparent authenticity. More often than not they are just a bit off. And sometimes they are just gibberish. As is widely acknowledged, neural text generation as it stands today has a significant problem: driven as it is by information that is ultimately about language use, rather than directly about the real world, it roams untethered to the truth. While the output of such a process might be good enough for the presidential teleprompter, it would not cut it if you want the hard facts about how your pension fund is performing. So, at least for the time being, nobody who develops commercial applications of NLG technology is going to rely on this particular form of that technology.
He give a list of commercial providers and concludes that most of them just offer smart templates. This is the important part:
> To the extent that you can tell from the clues to functionality that are surfaced by these various products, all the tools are ultimately very similar in terms of how they work, which might be referred to as ‘smart template’ mechanisms. There is a recognition that, at least for the kinds of use cases we see today, much of the text in any given output can be predetermined and provided as boilerplate, with gaps to be filled dynamically based on per-record variations in the underlying data source. Add conditional inclusion of text components and maybe some kind of looping control construct, and the resulting NLG toolkit, as in the case of humans and chimpanzees, shares 99% of its DNA with the legal document automation and assembly tools of the 1990s, like HotDocs (https://www.hotdocs.com). As far as I can tell, linguistic knowledge, and other refined ingredients of the NLG systems built in research laboratories, is sparse and generally limited to morphology for number agreement (one stock dropped in value vs. three stocks dropped in value).
That sounds pretty negative, but he emphasizes that putting an easy-to-use UI on well understood technology is meeting real business needs.
At the end he briefly touches on GPT2 and related technology.
I know plenty of closed-source tools that provide their authors significant income while benefiting from a healthy relationship with open source. Pico-8, which has a strong community of people who share game code, and Prodigy, which closely integrates with the open-source spaCy, both come to mind.
The spaCy developers' piece on running an open source business from 2017 is a much better approach to the topic.
This is another paper that's basically just about some details of word2vec and GloVe and their effects on the results:
Improving Distributional Similarity with Lessons Learned from Word Embeddings - ACL Anthology https://www.aclweb.org/anthology/Q15-1016/
1. The original word2vec is considered a reference implementation, used for benchmarks, so some people might not want it to change much.
2. The original hasn't been updated since 2017, and development was never very interactive.
3. word2vec has been widely reimplemented, and some reimplementations may be more widely used than the original (particularly Gensim).
4. The original had a paper from the very start (rather than being an open source project where a paper came later), so other papers reference it. For future papers that use the koan form, having a paper makes it easy to use in a similar way.
> Although continuous bag of word (CBOW) embeddings can be trained more quickly than skipgram (SG) embeddings, it is a common belief that SG embeddings tend to perform better in practice. This was observed by the original authors of Word2Vec [1] and also in subsequent work [2]. However, we found that popular implementations of word2vec with negative sampling such as word2vec and gensim do not implement the CBOW update correctly, thus potentially leading to misconceptions about the performance of CBOW embeddings when trained correctly.
The upshot is that they get similar results with CBOW while training three times faster than skipgram.
Given the popularity of Transformers, and that Fasttext exists, I'm curious as to what inspired them to even try this, but it's certainly an interesting result. There's so much word vector research that relies on quirks of the word2vec implementation.
I've had clients mention seeing my website or seeing my name in articles elsewhere, but I don't remember any mentioning HN. I've also gotten no reaction the few times I've posted on the hirings posts, even with an excellent endorsement one time. On the other hand people do find my site through search engines and I assume HN links help boost rankings there.
I have never used this and didn't realize it was possible, but I don't drive and rarely take taxis. I can see why this would be popular, especially with tourists - the language isn't a barrier to conveying a number. It also avoids issues with address formatting.
That said, two-number adresses are pretty common, I'm surprised you run into trouble with that. I live near Tokyo Tower, and even in the heart of Tokyo Minato-ku is only 99.XX% three-digit addresses. Azabu Nagasaka and Azabu Mamiana are tiny neighborhoods that are both still two digits (and Mamiana's name is another weird story...).
House numbers are usually of the form 1-2-3. Places with just two numbers are common, they haven't caught up with the "new" law from the 1962 yet.
Numbers are often written without suffixes, but in most places with three numbers the suffixes are 丁目-番-号. In one city - Wakasa in Fukui - the last two are reversed. I have never been able to find any mention of this at all, let alone why it would be the case.
Some cities are laid out in a clock pattern with slices named after animals of the Chinese Zodiac.
Neighborhoods are typically divided into blocks with numbers (or the number is omitted if there's just one block). But as addresses change over time, sometimes just one block goes away, so there are many places where block 1 no longer exists or there are gaps in the numbering.
On Yahoo Answers I once saw a person who was confused after their local post office adamantly told them one of their two or three numbers had a hyphen in it (so it was like 1-(2-3)) because it represented two joined lots, but that shouldn't be possible.
Mail is always complicated... I worked at a mail store in high school in the States and remember shipping some weird stuff.
https://www.wdic.org/w/CUL/%E8%86%A4 https://www.dampfkraft.com/ghost-characters.html