Four Short Links

Nat Torkington's eclectic collection of curated links.

Four short links: 21 July 2017

Offline First, Security Tools, Learning Game Strategy, and Design Documentation

  1. Offline First -- how to build an offline-first site in Javascript.
  2. Blackhat Arsenal -- software being released or updated during the Blackhat Arsenal event (e.g., DefPloreX, a machine-learning toolkit for large-scale e-crime forensics; and CBM, the "Car Backdoor Maker").
  3. Learning Macromanagement in StarCraft from Replays using Deep Learning -- Neural networks are trained on 789,571 state-action pairs extracted from 2,005 replays of highly skilled players, achieving top-1 and top-3 error rates of 54.6% and 22.9% in predicting the next build action. By integrating the trained network into UAlbertaBot, an open source StarCraft bot, the system can significantly outperform the game’s built-in Terran bot and play competitively against UAlbertaBot with a fixed rush strategy. (via Mark Riedl)
  4. Making Engineering Team Communication Clearer, Faster, Better -- it’s very important to make sure you have a process that actually gets people to read the document. The write-only document fired off into the void is a common problem, and this talks about how to solve it (for design documents, but the principles translate).

Four short links: 20 July 2017

SQL Equivalence, Streaming Royalties, Open Source Publishing, and Serial Entitlement

  1. Introducing Cosette -- a SQL solver for automatically checking semantic equivalences of SQL queries. With Cosette, one can easily verify the correctness of SQL rewrite rules, find errors in buggy SQL rewrites, build auto-graders for SQL assignments, develop SQL optimizers, bust “fake SQLs,” etc. Open source, from the University of Washington.
  2. Streaming Services Royalty Rates Compared (Information is Beautiful) -- the lesson is that it's more profitable to work for a streaming service than to be an artist hosted on it.
  3. Editoria -- open source web-based, end-to-end, authoring, editing, and workflow tool that presses and library publishers can leverage to create modern, format-flexible, standards-compliant, book-length works. Funded by the Mellon Foundation, Editoria is a project of the University of California Press and the California Digital Library.
  4. The Al Capone Theory of Sexual Harassment (Val Aurora) -- The U.S. government recognized a pattern in the Al Capone case: smuggling goods was a crime often paired with failing to pay taxes on the proceeds of the smuggling. We noticed a similar pattern in reports of sexual harassment and assault: often people who engage in sexually predatory behavior also faked expense reports, plagiarized writing, or stole credit for other people’s work.

Four short links: 19 July 2017

Open Source Car Code, Glass for Business, Videogame Narrative Skills, and Gmail Leverage

  1. Apollo -- open source autonomous auto platform.
  2. Glass for Enterprise -- Google X has relaunched Glass for businesses. See blog post and Steven Levy. A HUD for assembly operators in factories with marked results. “We knew the value of wearable technology when we first put it on the floor,” Gulick says. “In our first test in quality, our numbers were so high in the value it was adding that we actually retested and retested and retested. Some of the numbers we couldn't even publish because the leadership said they looked way too high.” I've been telling people for years that the killer app is boring business. Interesting that G has a new sales model: they make the hardware but sell to partners who will create specific applications and sell to customers.
  3. Game Writing: Narrative Skills for Videogames -- as VR and AR introduce branching stories into our lives (never mind Westworld-like hotels), the art of flexible narrative is a useful one to study. This is a review of a 2008 book with contributions from greats in the field. Of the books on professional games writing I’ve encountered, this is possibly the best, and definitely in the top three.
  4. Google Hire -- recruitment tool, but notable because they're integrating their enterprise tools into Gmail (note: you can't integrate your enterprise tool into Gmail). Most workflows touch email at some point, and that's the precise point where systems like Salesforce chafe. (I still twitch remembering the Chrome plugin for Gmail-Salesforce integration that we used at a previous startup.) Owning the mail client gives them huge opportunity here.

Four short links: 18 July 2017

Fooling Image Recognition, Electronics Text, Zero-Knowledge Proofs, and Massively Parallel Protein Design

  1. Robust Adversarial Inputs -- tricking deep learning image recognition models. We’ve created images that reliably fool neural network classifiers when viewed from varied scales and perspectives.
  2. CircuitLab Textbook -- free introductory electronics textbook, work in progress.
  3. The Hunting of the Snark -- a treasure hunt consisting of cryptographic challenges that will guide you through a zero-knowledge proof (ZKP) learning experience. As a reminder, zero-knowledge proofs, invented decades ago, allow verifiers to validate a computation on private data by allowing a prover to generate a cryptographic proof that asserts to the correctness of the computed output.
  4. Massively Parallel Protein Design -- We combined computational protein design, next-generation gene synthesis, and a high-throughput protease susceptibility assay to measure folding and stability for more than 15,000 de novo designed miniproteins, 1,000 natural proteins, 10,000 point mutants, and 30,000 negative control sequences. This analysis identified more than 2,500 stable designed proteins in four basic folds—a number sufficient to enable us to systematically examine how sequence determines folding and stability in uncharted protein space. Clever approach to understanding protein folding. (via Ian Haydon)

Four short links: 17 July 2017

Discarded GPUs, Go REPL, Learning Point Clouds, and 3D-Printing Nanopatterns

  1. Used GPUs Flood the Market as Ethereum Price Drops Below 150 -- On second-hand sales websites like eBay and Gumtree, we have seen a lot of new GPU listings appear in recent days, with plenty of used AMD RX series GPUs appearing over the weekend. More hardware is expected to hit these sites over the coming days as some miners wind down their operations, though many will simply move to a more profitable currency or to invest their computing power into an emerging cryptocurrency that has the prospect of high values in the future. That said, one HN commenter points out that in many areas with cheap power, it's still profitable to mine.
  2. go-pry -- An interactive REPL for Go that allows you to drop into your code at any point.
  3. Representation Learning and Adversarial Generation of 3D Point Clouds -- The expressive power of our learned embedding, obtained without human supervision, enables basic shape editing applications via simple algebraic manipulations, such as semantic part editing and shape interpolation. Figure 4 is the wow shot: interpolating between different tables, lounges, and chairs. (via Gene Kogan)
  4. Programming 2D/3D Shape-shifting with Hobbyist 3D Printers -- Here we present initially flat constructs that, upon triggering by high temperatures, change their shape to a pre-programmed 3D shape, thereby enabling the combination of surface-related functionalities with complex 3D shapes. Origami-like magic lets you print precisely controlled bio-nanopatterns, printed electronic components, and sensors/actuators.

Four short links: 14 July 2017

Molecular Sensing, Faking Speech, Radical Technologies, and Bullshit Detection

  1. Scio -- handheld molecular sensing for $300.
  2. AI Can Fake Speech (IEEE) -- The research team had a neural net analyze millions of frames of video to determine how elements of Obama's face moved as he talked, such as his lips and teeth and wrinkles around his mouth and chin. [...] In the new study, the neural net learned what mouth shapes were linked to various sounds. The researchers took audio clips and dubbed them over the original sound files of a video. They next took mouth shapes that matched the new audio clips and grafted and blended them onto the video. Essentially, the researchers synthesized videos where Obama lip-synched words he said up to decades beforehand.
  3. Radical Technologies: The Design of Everyday Life (Adam Greenfield) -- none of our instincts will guide us in our approach to the next normal. If we want to understand the radical technologies all around us, and see just how they interact to produce the condition we recognize as everyday life, we'll need a manual. That is the project of this book.
  4. Introductory Bullshit Detection for Non-Technical Managers -- “I’m creating a framework to...” It means: I’m not interested in solving the actual problem, so I’m going to create something else so that the person who actually will solve the problem has to also fix the problems in my stuff on top of that.

Four short links: 13 July 2017

Conversational Data Science, L3 Autonomy, Human Computation, and Embedded Learning

  1. Iris: A Conversational Agent for Data Science -- a cross between R Notebook and Facebook Messenger. See also this description of the project and what they hope to achieve.
  2. Audi A8: First to Reach Level 3 Autonomy -- for those of you not up with your autonomous driving levels, the A8 features the “AI traffic jam pilot,” meaning the car can take control of the driving in slow-moving traffic at up to 60 kilometers per hour. The system is activated by a button on the center console, and it can take over acceleration, braking, steering, and starting from a dead-stop, all without the driver paying attention.
  3. The Complexity of Human Computation: A Concrete Model with Application to Passwords -- The intent of this paper is to apply the ideas and methods of theoretical computer science to better understand what humans can compute in their heads. For example, can a person compute a function in their head so that an eavesdropper with a powerful computer—who sees the responses to random inputs—still cannot infer responses to new inputs?
  4. ELL -- Microsoft's Embedded Learning Library, which allows you to build and deploy machine-learned pipelines onto embedded platforms, like Raspberry Pis, Arduinos, micro:bits, and other microcontrollers. The deployed machine learning model runs on the device, disconnected from the cloud. Our APIs can be used either from C++ or Python.

Four short links: 12 July 2017

Net Neutrality, Scaling with Erlang, Matrix Cookbook, API Security

  1. Internet-Wide Day of Action to Save Net Neutrality -- light up the internet to fight the slashing of net neutrality laws; O'Reilly is part of it.
  2. How Discord Scaled Elixir -- the endless search for hot paths and bottlenecks that is scale.
  3. The Matrix Cookbook -- a collection of facts (identities, approximations, inequalities, relations, ...) about matrices and matters relating to them. It is collected in this form for the convenience of anyone who wants a quick desktop reference.
  4. API Security Checklist for Developers -- Checklist of the most important security countermeasures when designing, testing, and releasing your API.

Four short links: 11 July 2017

Brain Residency, Future of Work, Customer Feedback, and Verifying Distributed Systems

  1. The Google Brain Residency -- summary of outputs, with pointers to papers and GitHub repos, from the Google program that helps individuals from diverse educational backgrounds and experiences to dive into research in machine learning and deep learning.
  2. Findings of The Shift Commission on Work, Workers, and Technology -- their exploration of four scenarios. Rock-Paper-Scissors Economy: Less work, mostly tasks. Jump Rope Economy: More work, mostly tasks. King of the Castle Economy: Less work, mostly jobs. Go Economy: More work, mostly jobs. I like the two axis scenario framework: less/more work; jobs or tasks.
  3. Spotlight Feedback Framework -- depending on the questions your customers ask, you can categorize your problem as user experience ("how do I X?", "what happens when X?", "I tried to X"); product marketing ("can you/I X?", "how do you compare to X?", "how are you different than X?", "why should I use you for X?"); and positioning ("I'm probably not your target customer...", "I'm sure I'm wrong, but I thought..."). That distinction between "can I..." and "how do I..." is subtle; the former means you've not shared that your product can do the thing, while the latter means you've not made the method discoverable without explanation. (via Product Habits)
  4. Verdi -- Verdi is a framework from the University of Washington to implement and formally verify distributed systems. Verdi supports several different fault models, ranging from idealistic to realistic. Interesting on two fronts: proving system correctness may lead to useful tools and software development systems, and work on this to date has largely been around single-threaded programs. Anyone who has debugged multi-threaded code knows how much more difficult it is to reason about such systems.

Four short links: 10 July 2017

Multics Emulated, Physics Explained, DRM Vilified, and Lisp in Convenient Notebook Form​

  1. Running Multics on the DPS8M Emulator -- how-to guide that'll let you experience 1965 tech. I love that there's a new release of Multics (fore-runner of Unix) decades after the last hardware capable of running it died. (via Slashdot)
  2. The Mechanical Universe (YouTube) -- a critically acclaimed series of 52 30-minute videos covering the basic topics of an introductory university physics course. So, like a MOOC or Khan Academy, but from 1985. (via Caltech)
  3. DRM is Toxic to Culture -- can't be said enough. Technology-enforced restrictions quantize and prejudge discretion.
  4. Darkmatter -- The notebook-style Common Lisp environment​. ​

Four short links: 7 July 2017

Nanotube Computing, Civil Service Code, Ghastly Management, and Eminence-Induced Warpage

  1. Carbon Nanotube Computing -- The new 3D architecture is based on novel devices including two million carbon nanotube transistors and over one million resistive RAM cells, all built on top of a layer of silicon using existing fabrication methods and connected by densely packed metal wiring between the layers. As a demonstration, the team built an electronic nose that can sense and identify several common vapors, including lemon juice, rubbing alcohol, vodka, wine, and beer.
  2. Why I'm Leaving 18F -- 18F (and the larger service we created for it and its sibling organizations, the Technology Transformation Service), is being reorganized via administrative order into the General Services Administration’s (GSA) Federal Acquisition Service. [...] We were subsequently told that the new Commissioner of the Federal Acquisition Service would suddenly and immediately become a political position, with a person appointed directly by the White House. In a single day, the White House took direct control of two of the most important shared service organizations in government.
  3. Antisocial Coding: My Year at GitHub -- For my first few pull requests, I was getting feedback from literally dozens of engineers (all of whom were male) on other teams, nitpicking the code I had written. One PR actually had over 200 comments from 24 different individuals. It got to the point where the VP of engineering had to intervene to get people to back off. From there to an HR-orchestrated firing, in a year (via a pile of undercutting nonsense). Grim, and a textbook lesson in how not to treat your employees and coworkers. Our industry needs an enema.
  4. Our Obsession With Eminence Warps Research (Nature) -- cf "10x engineers."

Four short links: 6 July 2017

Mobile Testing, Semantic Segmentation, Salary Gossip, and Word Analogies

  1. OWASP Mobile Testing Guide -- a work in progress, but useful.
  2. Semantic Segmentation Using Deep Learning -- review of the techniques that help you categorize pixels in an image by which object in the image they belong to.
  3. Salary Gossip -- Top 6% or so of engineers at Amazon, Oracle, Google, Facebook, Twitter are paid more than $1.3 million per year. Next 11% make $650,000 on average. [...] $600,000 to $2 million packages are similarly becoming common in the U.S. Software engineers with 10 years experience should be making ~$420,000 per year with ~$210,000 salary. [...] $240,000 to $470,000 packages are now common in China. [...] Fresh MS graduates in the U.S. are getting $220,000+ packages at Google, Amazon, eBay, Twitter, LinkedIn, Airbnb, Facebook, Snapchat, will add more. No idea of the source, so take it all with a pillar of salt, but ... holy crap. As a Kiwi, I approve of Do not go to Singapore, Germany, or U.K. Go to Canada, Australia, France, New Zealand, or South Africa instead.
  4. Word Vectors and SAT Analogies -- clever approach using the SAT analogy questions to test how well the word vector technique ("king - man + woman = queen") holds up to relatively real-life situations. He got 49% accuracy but notes: These methods are not the best-performing non-human technique for these SAT analogy questions. Littman and Turney report several. Latent Relational Analysis comes in at 56% accuracy, against the average U.S. college applicant at 57%.

Four short links: 5 July 2017

Network Mathematics, Learning Math, Open Source Q&A, and Australian Privacy

  1. Network Mathematics and Rival Factions -- understanding how "the enemy of my enemy" plays out. (via Steven Strogatz)
  2. A Path Less Taken to the Peak of the Math World -- the ultimate in "fake it until you make it." Huh tried to use these lunches to ask Hironaka questions about himself, but the conversation kept coming back to math. When it did, Huh tried not to give away how little he knew. “Somehow I was very good at pretending to understand what he was saying,” Huh said. Indeed, Hironaka doesn’t remember ever being aware of his would-be pupil’s lack of formal training. “It’s not anything I have a strong memory of. He was quite impressive to me,” he said.
  3. Gitter Open Sourced -- source to GitLab's Q&A site about open source software. Useful, I guess, if you want to run your own Q&A site.
  4. Medicare Details of Every Australian Currently Up for Sale on the Dark Web -- this is fine.

Four short links: 4 July 2017

Retro Fonts, AI Research, Mind Reading, and Prototyping Tool

  1. Old-School PC Fonts -- I code better when my font is retro. My setup: Print Char 21 font, green on black. (via Hervé Piton)
  2. Measuring the Progress of AI Research (EFF) -- This pilot project collects problems and metrics/data sets from the AI research literature, and tracks progress on them. Some astonishing exponential growth with even more astonishing x-axes of time measured in months. (via Chris Dixon)
  3. Mind Reading Comes a Step Closer -- The model was able to predict the features of the left-out sentence with 87% accuracy, despite never being exposed to its activation before. It was also able to work in the other direction: to predict the activation pattern of a previously unseen sentence, knowing only its semantic features.
  4. Pencil -- tool for making diagrams and GUI prototyping that everyone can use. Free and open source.

Four short links: 3 July 2017

Robot Academy, Wikipedia Adventure, PR Science, and Complexity and Planning

  1. Robot Academy -- University level, short video lessons and full online courses to help you understand and prepare for this technology of the future. From the Queensland University of Technology.
  2. Wikipedia: The Text Adventure -- this is so clever!
  3. An Adversarial Review of Adversarial Generation of Natural Language - while I agree that short publication cycles on arXiv can be better than the lengthy peer-review process we now have, there is also a rising trend of people using arXiv for flag-planting, and to circumvent the peer-review process. This is especially true for work coming from “strong” groups. Currently, there is practically no downside of posting your (often very preliminary, often incomplete) work to arXiv, only potential benefits.
  4. The Critical Difference Between Complex and Complicated -- In a complex environment, it is truly rare that a grand plan or strategy will work as intended. (via Stuart Candy)

Four short links: 30 June 2017

Jumping Robot, Exactly Once, Pocket Negotiator, and Science-Based Games

  1. Salto, the Jumping Robot (IEEE Spectrum) -- two little thrusters are able to control Salto-1P’s yaw and roll: When they’re thrusting in different directions, the robot yaws, and when they both thrust in the same direction, the robot rolls. Combined with the tail, that means Salto-1P (which only ways 98 grams) can stabilize and control itself in three dimensions, even in mid-air, which is what allows it to chain together so many jumps.
  2. Delivering Billions of Messages Exactly Once -- using Kafka and RocksDB to build a "two-phase architecture" to give the commit and rollback needed.
  3. Pocket Negotiator -- TU Delft's software will collaborate with you in your upcoming negotiation to make it a pleasent experience ending in a good deal. The Pocket Negotiator can be used for preparation (training session), to support during actual negotiations and for mediating support.
  4. Science-Based Games List --a collaborative notepad with educational/science games—i.e., games that are: capturing parts of real scientific phenomena (including social science, medicine, etc.), and actually playable (you can play them for fun, not ones "for classroom only").

Four short links: 29 June 2017

Bug Chasing, Brain Videos, Collaborative Annotation, and TLD Scam

  1. A Bug Detective Story -- spoiler: the CPU did it.
  2. How to Capture Videos of Brains in Real Time -- The scientists first engineered the animals’ neurons to fluoresce (glow), using a method called optogenetics. The stronger the neural signal, the brighter the cells shine. To capture this activity, they used a technique known as “light-field microscopy,” in which an array of lenses generates views from a variety of perspectives. These images are then combined to create a three-dimensional rendering, using a new algorithm called “seeded iterative demixing” (SID) developed by the team.
  3. Lacuna -- open source collaborative annotation for tertiary classes, built at Stanford.
  4. The .feedback Scam -- wow. An entire top-level domain dedicated to the old "someone is talking about you, sign up to find out what they're saying" scam.

Four short links: 28 June 2017

Robot Security, Deep Grasping, Augmenting Programmers, and Crypto Database Tech

  1. An Experimental Security Analysis of an Industrial Robot Controller -- it's a bloodbath. Useless authentication, trivial crypto, privilege escalation, and a security checker that isn't itself integrity checked. And more. (via Adrian Colyer)
  2. Dexterity Network (Dex-Net) 2.0 Dataset for Deep Grasping -- "deep grasping" sounds like the tagline for late-stage capitalism, but is actually applying deep learning to the robotics task of grabbing things. This Berkeley-released training data set consists of 6.7M sets of synthetic point clouds and depth information as a robot grasps an object, labeled with how successful the robot's grasp was. It enables researchers to build their own models of how successful grasps will be, so their robots can make better decisions than Berkeley's. Bless their grasping little paws.
  3. Diffblue Research -- applying AI to software creation. It's not just pop that will eat itself. Topics include: automated detection of relevant code features, making changes to code automatically, and designing techniques to find exploitable bugs.
  4. Cryptographically Protected Database Search -- we provide the following contributions: 1) An identification of the important primitive operations across database paradigms. We find there are a small number of base operations that can be used and combined to support a large number of database paradigms. 2) An evaluation of the current state of protected search systems in implementing these base operations. This evaluation describes the main approaches and tradeoffs for each base operation. Furthermore, it puts protected search in the context of unprotected search, identifying key gaps in functionality. 3) An analysis of attacks against protected search for different base queries. 4) A roadmap and tools for transforming a protected search system into a protected database, including an open source performance evaluation platform and initial user opinions of protected search. (via Adrian Colyer)

Four short links: 27 June 2017

Reading Papers, AR Kit, Demoing to Sell, and Secure Go

  1. How to Read a Scientific Paper: A Guide for Non-Scientists -- excellent advice. 1. Read the introduction (not the abstract). 2. Identify the BIG QUESTION. 3. Summarize the background in five sentences. 4. Identify the SPECIFIC QUESTION(S). 5. Identify the approach. 6. Draw a diagram for experiments, showing methods. 7. Summarize results from each experiment. 8. Do results answer the SPECIFIC QUESTION(S). 9. Read conclusion/discussion/interpretation section. 10. Now, read the abstract. 11. What do other researchers say about this paper?
  2. Made with ARKit -- selections of demos made with Apple's augmented reality framework. I may be shallow, but I'm excited to have this in my hands. The ruler made me go "wow."
  3. Everything I Wish I'd Known Before I Started Demoing SaaS -- the sales process turns on the pain points and the decision-maker, and this (good!) advice is how you make sure your demo moves you ahead on both.
  4. Go Language - Web Application Secure Coding Practices -- The main goal of this book is to help developers avoid common mistakes, while at the same time learning a new programming language through a "hands-on approach." This book provides a good level of detail on "how to do it securely," showing what kind of security problems could arise during development. (via Binni Shah)

Four short links: 26 June 2017

Howtoons Kidsets, Computational Thinking, Data Viz, and Affordable Genomics

  1. Howtoons Subscription -- this is brilliant! An amazing kit in the mail each month, with an interesting project each time. It's given me some priceless moments with my nephew. They're not so long that the kid loses interest, nor so mundane that you lose interest.
  2. CS Unplugged 2.0 -- an update to the classic "learn computational thinking without a computer" system. (via Jack Morgan)
  3. Data Visualization Pitfalls to Avoid (Tamara Munzner) -- latest iteration of this excellent evidence-based guide to making your visualizations accurate, useful, and generally free of the suck. If you're not periodically checking Tamara's talks page, then you're missing out.
  4. Oxford Nanopore -- I've been watching this for a while, and it's showing all the signs of enabling the promised genomics explosion: affordable real-time USB-powered DNA sequencing. Users are doing all sorts of interesting things from amateur soil metagenomics to real-time Zika sequencing.