Top all-time · #2176
State of the Smerity
Personal blog · smerity.com/articles · on HN as Smerityaiprogramming
State of the Smerity is a personal blog at smerity.com about ai and programming. Since 2009, 34 different Hacker News readers have linked it 71 times, most often “Stop saying DeepCoder steals code from StackOverflow”. No feed was found, so read it on the site itself.
Too few sampled comments (11) to say how often HN recommends it.?
Why HN links it
HN readers link Smerity's articles as explainers on machine learning, such as a neural machine translation tutorial and a Keras QA write-up, plus pieces on web tracking, Amazon's security and Tay.
- Recommended for · 4 comments
- Called interesting knowledge, a good primer on the basics of the game, a nice visual tutorial on NMT, and worth reading before jumping to conclusions about Tay.
Summarised by Claude from a sample of 11 HN comments that link here. Each comment is marked recommending, critical or neutral (citing without judging); how this works.
An increasingly public process journal of an independent AI researcher
Most-linked posts?
- Stop saying DeepCoder steals code from StackOverflow5 people · 5 mentions · first linked 2017-02-27
- Peeking into the architecture used for Google's Neural Machine Translation4 people · 7 mentions · first linked 2016-11-18
- In deep learning, architecture engineering is the new feature engineering4 people · 4 mentions · first linked 2016-06-13
- Question answering on the Facebook bAbi dataset using RNN3 people · 4 mentions · first linked 2015-08-06
- Where did all the HTTP referrers go?3 people · 3 mentions · first linked 2013-05-28
- Simple questions to help reduce AI hype2 people · 4 mentions · first linked 2016-07-08
- TayAndYou – toxic before human contact2 people · 4 mentions · first linked 2016-03-25
- What to Pay Attention to in the OpenAI Five Benchmark2 people · 2 mentions · first linked 2018-08-05
- /articles/2016/google_nmt_arch.html2 people · 2 mentions · first linked 2017-01-06
- The lack of HTTPS at Amazon: identifying items purchased via information leakage2 people · 2 mentions · first linked 2015-04-04
What Hacker News says?
This is interesting knowledge: https://smerity.com/articles/2013/where_did_all_the_http_ref...
Comment linking /articles/2013/where_did_all_the_http_referrers_go.html
on “How I recorded user behaviour on my competitor’s websites”Submitted Simple questions to help reduce AI hype as a story · 208 points
HN discussionTo expand on this a bit: You need a baseline feature representation for structure data for "traditional" problems. What it DOES is shifts the problem to layer architectures which is a hard problem in and of itself: What vonnik is talking about and what's already been addressed in a sibling comment was images. "Features" are the pixels which handles generating the "baseline representation" needed for feature engineer…
Comment linking /articles/2016/architectures_are_the_new_feature_engineering.html
on “Approaching Almost Any Machine Learning Problem”Submitted The lack of HTTPS at Amazon: identifying items purchased via information leakage as a story · 128 points
HN discussionI'm glad this is happening and honestly surprised this didn't happen earlier. fchollet, the Keras main author, works at Google and had integrated TensorFlow into Keras very early on (given he'd known about the existence long before the general public). Even though Keras is backend agnostic, you don't really lose any of the flexibility that TensorFlow would give you - it's a pretty transparent abstraction[3]. Best of…
Comment linking /articles/2016/orthogonal_init.html
on “Keras will be added to core TensorFlow at Google”Submitted Where did all the HTTP referrers go? as a story · 110 points
HN discussionGood rule-of-thumb to always add dropout. Very hard to lower the performance this way. Yes, we have to try a load of stuff, especially in deep learning, where the old feature engineering seems to be replaced with architecture engineering [1]. Yet, the stuff to try is usually well-known [2]. You should not be trying stuff completely at random (unless you do a random gridsearch for hyperparams [3]). > we chose 3 layer…
Comment linking /articles/2016/architectures_are_the_new_feature_engineering.html
on “Why is machine learning ‘hard’?”Submitted Where did all the HTTP referrers go? (2013) as a story · 97 points
HN discussionI really appreciated this story. It's so hard to make people care about "small" data leaks when they have no idea what many "small" data leaks can lead to. The bug the journalists discovered that revealed Skype's contact list is the perfect example -- the programmer just did something completely "reasonable" (grabbing avatars) that ended up leaking a vital set of information. Imagine if surveillance was your full ti…
Comment linking /articles/2013/google_analytics_and_nsa.html
on “Ars tests Internet surveillance by spying on an NPR reporter”Submitted The compute and data moats are dead as a story · 82 points
HN discussionYou hit the nail on the head. Google Analytics is on a substantial proportion of the Internet. 65% of the top 10k sites, 63.9% of the top 100k, and 50.5% of the top million[1]. My own results from a research project I did using the Common Crawl[2] corpus estimates approximately 39.7% of the 535 million pages processed so far have GA on them. The real key to tracking is the referrer data. For the vast majority of cli…
Comment linking /articles/2013/google_analytics_and_nsa.html
on “Google Has Most of My Email Because It Has All of Yours”Submitted Peeking into the architecture used for Google's Neural Machine Translation as a story · 81 points
HN discussionI did a light benchmark of their 512MB VPS[1] which is insanely cheap. The only things that change when moving to the 2GB VPS is RAM, hard drive space and total data transfer. The CPU remains a single core and the same speed according to [2]. It's possible there's less contention but you'd need to find a benchmark for the 2GB VPS to know for sure. If you're after any specific benchmarks give me a yell and I'll see i…
Comment linking /articles/2012/hetzner.html
on “European server provider Hetzner now offers a 64GB RAM box for 109€/mon”Submitted Google, make Google Analytics HTTPS by default as a story · 75 points
HN discussionA good thing to read to know the basics of the game and what to pay attention to: http://smerity.com/articles/2018/n_things_to_look_out_for_in...
Comment linking /articles/2018/n_things_to_look_out_for_in_openai_benchmark.html
on “OpenAI bots competing against Humans right now”Submitted The Mystery of the Spotty Animated GIF as a story · 65 points
HN discussion>Almost every recent A.I. advance has come from one tiny corner of the field, machine learning. Machine learning exposes a set of connected nodes, known as neural nets, to mass amounts of labeled real-world data in an attempt to give those nodes tacit knowledge. The breakthrough example was software that was able to identify cat pictures. >So far, these neural nets have given us some great demos but mostly niche rea…
Comment linking /articles/2016/google_nmt_arch.html
on “Why Doesn't Software Show Up in Productivity?”Submitted Measuring the impact of Google Analytics (2013) as a story · 60 points
HN discussionIf people are interested in the underlying architecture of Google's Neural Machine Translation (GNMT) system, I wrote an article that builds it up piece by piece. While it's intended for people who are likely to implement GNMT or similar architectures, the article is descriptive enough that it should be possible to follow along even if you're not well versed in deep learning. http://smerity.com/articles/2016/google_…
Comment linking /articles/2016/google_nmt_arch.html
on “Zero-Shot Translation with Multilingual Neural Machine Translation System”Submitted How Google Sparsehash achieves two bits of overhead per entry using sparsetable as a story · 54 points
HN discussionConsidering the importance of HTTPS to, in Google's words, "[making the] Internet safer more broadly", this seems like a good time to again suggest that Google enable HTTPS for Google Analytics by default[1]. Google Analytics is on 50.8% of the top million domains on the Internet, and on 26.96% of a randomly selected 48.5 million domains[1]. Of the 42 billion links analyzed in my research, over 48% of them had Googl…
Comment linking /articles/2013/google_analytics_and_nsa.html
on “HTTPS as a ranking signal”Submitted TayAndYou – toxic before human contact as a story · 24 points
HN discussionThe diagram was generated using a DMN I've implemented in TensorFlow, part of my work at MetaMind. Those diagrams are useful not just in visualizing the architecture but also for spot checking certain issues. I was planning on writing up a blog post on TensorFlow in the near future but undecided on the topic. It could be about implementing something nice and simple[1], maybe an attention based model / language model…
Comment linking /articles/2015/keras_qa.html
on “Running your models in production with TensorFlow Serving”Submitted Question answering on the Facebook bAbi dataset using RNN as a story · 12 points
HN discussionIt honestly disturbs me that Myspace, Justin.tv, and these other companies don't seem to ever consider shipping their soon-to-be-deceased data to the Internet Archive. Why not help preservation a little? It'd likely be simpler to handle than the oncoming denial of preservation defense that the ArchiveTeam launches. Have there been any examples of companies doing this? You might think that this is all useless informa…
Comment linking /articles/2013/deleted_digital_tombstones.html
on “Justin.tv ending archiving, deleting all archived videos after June 14”Submitted In deep learning, architecture engineering is the new feature engineering as a story · 8 points
HN discussionIn practice, images are not particularly large and a batch of them would easily fit on a single GPU. What's more common is either (a) performing the forward and backward passes on 4 GPUs where each GPU has its own batch, then collecting the gradient from all 4 backward passes or (b) splitting the computation for individual layers across multiple GPUs. Both (a) and (b) have various trade-offs. Some models perform wor…
Comment linking /articles/2016/google_nmt_arch.html
on “DeepLearning10: The 8x Nvidia GTX 1080 Ti GPU Monster”Submitted Stop saying DeepCoder steals code from StackOverflow as a story · 6 points
HN discussionIf people are interested in getting started with question answering on the bAbi dataset, I wrote a baseline RNN example that's now in Keras[1]. Whilst I love the theory behind the bAbi dataset, there are issues with the data itself that still need to be fixed, particularly that there is duplication in the datasets[2]. This duplication leads to less unique data than expected and even overlaps between training and tes…
Comment linking /articles/2015/keras_qa.html
on “Towards Neural Network-based Reasoning”Submitted Cloudflare's Cloudbleed in plain English as a story · 6 points
HN discussion
