Archive of posts with tag 'Machine Learning'

Lexica.art →

October 11, 2022 • #

The community around Stable Diffusion, the open source AI project for text-to-image generation, has been buzzing. From nonexistent a year ago to thousands of contributors and forks and spinoffs. There’s even a GUI macOS app.

Lexica is a project to index and make prompts and images from Stable Diffusion searchable. Playing around with it, it’s pretty impressive. So much incredible possibility here. This tech will make the volume of content on the internet literally infinite.

Stable Diffusion images for 'architecture'

✦

Weekend Reading: Intellectual Humility, Scoping, and Gboard

August 31, 2019 • #

🛤 Missing the Light at the End of the Tunnel

Honest postmortems are insightful to get the inside backstory on what happened behind the scenes with a company. In this one, Jason Crawford goes into what went wrong with Fieldbook before they shut it down and were acquired by Flexport a couple years ago:

Now, with a year to digest, I think this is true and was a core mistake. I vastly underestimated the resources it was going to take—in time, effort and money—to build a launchable product in the space.

In the 8 years since we launched the first version of Fulcrum, we’ve had (fortunately) smaller versions of this experience over and over. Each new major overhaul, large feature, or product business model change we’ve undertaken has probably cost us twice the time we initially expected it to. Scoping is a science itself that everyone has to learn.

In Jeff Bezos’s 2018 letter to Amazon shareholders, he discusses the topic of high standards: how to have them and how to get your team to have them. (As a side note, if you don’t read Bezos’s shareholder letters, you’re missing out. Even if you’ve already read all the business and startup advice in the world, you will find new and keen insights there.)

Bezos makes a few interesting points, but I’ll focus on one: To have high standards in practice, you need realistic expectations about the scope of effort required.

As a simple example, he mentions learning to do a handstand. Some people think they should be able to learn a handstand in two weeks; in reality, it takes six months. If you go in thinking it will take two weeks, not only do you not learn it in two weeks, you also don’t learn it in six months—you learn it never, because you get discouraged and quit. Bezos says a similar thing applies to the famous six-page memos that substitute for slide decks at Amazon (the ones that are read silently in meetings). Some people expect they can write a good memo the night before the meeting; in reality, you have to start a week before, in order to allow time for drafting, feedback, and editing.

🏛 Ten Ways to Defuse Political Arrogance

David Blankenhorn calls for a return of intellectual humility in public discourse.

At the personal level, intellectual humility counterbalances narcissism, self-centeredness, pridefulness, and the need to dominate others. Conversely, intellectual humility seems to correlate positively with empathy, responsiveness to reasons, the ability to acknowledge what one owes (including intellectually) to others, and the moral capacity for equal regard of others. Arguably its ultimate fruit is a more accurate understanding of oneself and one’s capacities. Intellectual humility also appears frequently to correlate positively with successful leadership (due especially to the link between intellectual humility and trustworthiness) and with rightly earned self-confidence.

⌨️ The Machine Intelligence Behind Gboard

A fun technical overview of how the Google team is using predictive machine learning models to make typing on mobile devices more efficient.

✦

Weekend Reading: Universal Laws, Tandem, and Computers That Can See

August 3, 2019 • #

📚 Universal Laws of the World

A list of broad laws that apply to all fields. Thoughtful stuff as always from Morgan Housel:

6. Parkinson’s Law: Work expands to fill the time available for its completion.

In 1955 historian Cyril Parkinson wrote in The Economist:

IT is a commonplace observation that work expands so as to fill the time available for its completion. Thus, an elderly lady of leisure can spend the entire day in writing and despatching a postcard to her niece at Bognor Regis. An hour will be spent in finding the postcard, another in hunting for spectacles, half-an-hour in a search for the address, an hour and a quarter in composition, and twenty minutes in deciding whether or not to take an umbrella when going to the pillar-box in the next street. The total effort which would occupy a busy man for three minutes all told may in this fashion leave another person prostrate after a day of doubt, anxiety and toil.

His point was that resources can exceed needs without people noticing. The number of employees in an organization is not necessarily related to the amount of work that needs to be done in that organization. Workers will find something to do – or the appearance of doing something – regardless of what needs to be done.

💬 Tandem

This is a neat collaboration tool for distributed teams that just launched. It’s built on Slack and has integrations built for many of the common productivity tools that modern remote teams are familiar with. I’m keen to take a look at this for doing more real-time work with my remote co-workers.

👁 Computers That Can See

As computer vision continues its advance, machines are getting better and better at converting images and video into structured data. Computers have historically had sensor data feeds through text, binary data streams, and user inputs; eventually they’ll all have visual inputs, as well.

✦

Assisted Tasks for OpenStreetMap →

May 30, 2019 • #

The Humanitarian OpenStreetMap Team has been working on an experimental version of the Tasking Manager tool that incorporates deep learning-assisted mapping projects.

The OSM community has long been (and still largely is) averse to machine-based mapping, as it’s counter to the founding ethos of the project being “created by mappers, where they live.” But if the project is to survive and still see adoption and usage in commercial applications, there has to be effort to improve the depth of coverage and refresh rate to stay competitive with the commercial providers like Google and HERE. They’re out robotically and mechanically mapping the world trying to actively remove humans from the loop.

It’s good to see HOT taking after work done already by Facebook, Mapbox, and Development Seed to combine ML capabilities with OSM to improve the map at scale.

✦

Weekend Reading: Data Moats, China, and Distributed Work

May 25, 2019 • #

🏰 The Empty Promise of Data Moats

In the era of every company trying to play in machine learning and AI technology, I thought this was a refreshing perspective on data as a defensible element of a competitive moat. There’s some good stuff here in clarifying the distinction between network effects and scale effects:

But for enterprise startups — which is where we focus — we now wonder if there’s practical evidence of data network effects at all. Moreover, we suspect that even the more straightforward data scale effect has limited value as a defensive strategy for many companies. This isn’t just an academic question: It has important implications for where founders invest their time and resources. If you’re a startup that assumes the data you’re collecting equals a durable moat, then you might underinvest in the other areas that actually do increase the defensibility of your business long term (verticalization, go-to-market dominance, post-sales account control, the winning brand, etc).

Companies should perhaps be less enamored of the “shiny object” of derivative data and AI, and instead invest in execution in areas challenging for all businesses.

🇨🇳 China, Leverage, and Values

An insightful piece this week from Ben Thompson on the current state of the trade standoff between the US and China, and the blocking of Chinese behemoths like Huawei and ZTE. The restrictions on Huawei will mean some major shifts in trade dynamics for advanced components, chip designs, and importantly, software like Android:

The reality is that China is still relatively far behind when it comes to the manufacture of most advanced components, and very far behind when it comes to both advanced processing chips and also the equipment that goes into designing and fabricating them. Yes, Huawei has its own system-on-a-chip, but it is a relatively bog-standard ARM design that even then relies heavily on U.S. software. China may very well be committed to becoming technologically independent, but that is an effort that will take years.

The piece references this article from Bloomberg, an excellent read on the state of affairs here.

⌨️ The Distributed Workplace

I continue to be interested in where the world is headed with remote work. Here InVision’s Mark Frein looks back at what traits make for effective distributed companies, starting with history of past experiences of remote collaboration from music production, to gaming, to startups. As he points out, you can have healthy or harmful cultures in both local and distributed companies:

Distributed workplaces will not be an “answer” to workplace woes. There will be dreary and sad distributed workplaces and engaged and alive ones, all due to the cultural experience of those virtual communities. The key to unlocking great distributed work is, quite simply, the key to unlocking great human relationships — struggling together in positive ways, learning together, playing together, experiencing together, creating together, being emotional together, and solving problems together. We’ve actually been experimenting with all these forms of life remote for at least 20 years at massive scales.

✦

Weekend Reading: LiDAR, Auto Generated Textbooks, and Paleo Plate Tectonics

February 9, 2019 • #

🛣 Creating Low-Cost LiDAR

This is a great breakdown of the different elements of LiDAR technology, looking at three broad areas: beam direction, distance measurement, and frequencies. They compare the tech of 10 different companies in the space to see how each is approaching the problem.

📚 An Algorithm to Auto-Generate Textbooks

Taking off of the Wikibooks project, this team is aiming to generate books from Wikipedia content using ML techniques.

Given the advances in artificial intelligence in recent years, is there a way to automatically edit Wikipedia content so as to create a coherent whole that is useful as a textbook? Enter Shahar Admati and colleagues at the Ben-Gurion University of the Negev in Israel. These guys have developed a way to automatically generate Wikibooks using machine learning. They call their machine the Wikibook-bot. “The novelty of our technique is that it is aimed at generating an entire Wikibook, without human involvement,” they say.

🌍 Paleogeographic History of Plate Tectonics

This simple app lets you slide from the Jurassic to the Holocene. A vivid demonstration of how long 200 million years really is.

Paleo Plate Tectonics

✦

A New Approach to Understanding How Machines Think →

January 22, 2019 • #

This is an interesting interview with Been Kim from Google Brain on developing systems for seeing how trained machines make decisions. One of the major challenges with neural network-based based deep learning systems is that the decision chain used by the AI is a black box to humans. It’s difficult (or impossible) for even the creators to figure out what factors influenced a decision, and how the AI “weighted” the inputs. What Kim is developing is a “translation” framework for giving operators better insight into the decision chain of AI:

Kim and her colleagues at Google Brain recently developed a system called “Testing with Concept Activation Vectors” (TCAV), which she describes as a “translator for humans” that allows a user to ask a black box AI how much a specific, high-level concept has played into its reasoning. For example, if a machine-learning system has been trained to identify zebras in images, a person could use TCAV to determine how much weight the system gives to the concept of “stripes” when making a decision.

✦

The Incredible Inventions of Intuitive AI

January 2, 2019 • #

This talk on “generative AI” was interesting. One bit stuck out to me as really thought-provoking:

Dutch designers have created a system to 3D print functional things in-place, like this bridge concept. Imagine that you can place a machine, give it a feed of raw material input and cut it loose to generate something in physical space. As the presenter mentions at the end of the talk, moving from things that are “constructed” to ones that are “grown”.

✦

Amazon SageMaker Ground Truth →

November 30, 2018 • #

This week was Amazon’s annual re:Invent conference, where they release n + 10 new products for AWS (where n is the number of products launched at last year’s event). It’s mind-boggling how many new things they can ship each year.

SageMaker was launched last year as a platform for automating machine learning pipelines. One of the missing pieces was the ability to build training datasets with your own custom data. That’s the intent with Ground Truth. It supports building your dataset in S3 (like a group of images), creating a labeling task, and distributing it to a team to annotate to train a model. It integrates with Mechanical Turk, Amazon’s network of third-party vendors, or your own private team. This is awesome for anyone with massive datasets but no easy-to-use system to build the training info.

This, combined with their Rekognition product open up some interesting possibilities for image recognition use cases I’d like to test out.

✦

Weekly Links: Tensor Processing, Amazon, and Preventing Traffic Jams

April 13, 2017 • #

Google’s “Tensor Processing Unit” 💻

Google has built their own custom silicon dedicated to AI processing. The power efficiency gains with these dedicated chips is estimated to have saved them from building a dozen new datacenters.

But about six years ago, as the company embraced a new form of voice recognition on Android phones, its engineers worried that this network wasn’t nearly big enough. If each of the world’s Android phones used the new Google voice search for just three minutes a day, these engineers realized, the company would need twice as many data centers.

Jeff Bezos’ Annual Letter to Shareholders 📃

An excellent read. Their philosophy of experimentation comes through. I liked this bit, on the “velocity” of decision making:

Day 2 companies make high-quality decisions, but they make high-quality decisions slowly. To keep the energy and dynamism of Day 1, you have to somehow make high-quality, high-velocity decisions. Easy for start-ups and very challenging for large organizations. The senior team at Amazon is determined to keep our decision-making velocity high. Speed matters in business – plus a high-velocity decision making environment is more fun too. We don’t know all the answers, but here are some thoughts.

First, never use a one-size-fits-all decision-making process. Many decisions are reversible, two-way doors. Those decisions can use a light-weight process. For those, so what if you’re wrong? I wrote about this in more detail in last year’s letter.

Second, most decisions should probably be made with somewhere around 70% of the information you wish you had. If you wait for 90%, in most cases, you’re probably being slow. Plus, either way, you need to be good at quickly recognizing and correcting bad decisions. If you’re good at course correcting, being wrong may be less costly than you think, whereas being slow is going to be expensive for sure.

How not to create traffic jams, pollution and urban sprawl 🚘

The Economist analyzes the state of parking economics. The gist: free or low-cost parking equals congestion and more drivers roaming for longer. Some great statistics in this piece:

As San Francisco’s infuriated drivers cruise around, they crowd the roads and pollute the air. This is a widespread hidden cost of under-priced street parking. Mr. Shoup has estimated that cruising for spaces in Westwood village, in Los Angeles, amounts to 950,000 excess vehicle miles travelled per year. Westwood is tiny, with only 470 metered spaces.

✦

Browse the full archive →

✦