Google said Wednesday it’s making a handful of privacy changes for users.
In a blog post, the search giant said it’ll make it easier for users to go “incognito” and pause data collection as they use Google’s mobile apps. It’s also adding proactive account security recommendations to its Security Checkup tool.
Google also said it’ll allow its users to auto-delete their data after 18 months.
It explains:
Starting today, the first time you turn on Location History — which is off by default — your auto-delete option will be set to 18 months by default. Web & App Activity auto-delete will also default to 18 months for new accounts. This means your activity data will be automatically and continuously deleted after 18 months, rather than kept until you choose to delete it. You can always turn these settings off or change your auto-delete option.
Meanwhile stateside, Google is one of many Silicon Valley tech giants also facing renewed questions about how they reconcile protecting users’ privacy while still providing their technology to law enforcement, amid protests against police brutality in the wake of the killing of George Floyd.
Over a thousand of Google’s own employees have asked the company to stop selling its technology to police departments across the United States. Amazon, IBM and Microsoft have already pulled the plug on selling their facial recognition technology to police, but left a wide berth to still sell to federal law enforcement.
Google is also facing a $5 billion class action suit in California for allegedly pervasively tracking the internet use of its users through their browser’s “incognito” mode.
For decades, the United States has had a monopoly on positioning, navigation, and timing technology with its Global Positioning System (GPS), a constellation of satellites operated by the military that today provides the backbone for location on billion of devices worldwide.
As those technologies have become not just key to military maneuvers but the very foundation of modern economies, more and more governments around the world have sought ways to decouple from usage of the U.S.-centric system. Russia, Japan, India, the United Kingdom and the European Union have all made forays to build out alternatives to GPS, or at least, to augment the system with additional satellites for better coverage.
Few countries though have made the investment that China has made into its Beidou (北斗) GPS alternative. Over twenty years, the country has spent billions of dollars and launched nearly three dozen satellites to create a completely separate system for positioning. According to Chinese state media, nearly 70% of all Chinese handsets are capable of processing signals from Beidou satellites.
Now, the final puzzle piece is in place, as the last satellite in the Beidou constellation was launched Tuesday morning into orbit, according to the People’s Daily.
While the completion of the current plan for Beidou offers Beijing new flexibility and resiliency for this critical technology, ultimately, positioning technologies are mostly not adversarial — additional satellites can offer more redundancy to all users, and many of these technologies have the potential to coordinate with each other, offering more flexibility to handset manufacturers.
Given how much of global logistics and our daily lives are controlled by these technologies, further international cooperation around protecting these vital assets seems necessary. Now that China has its own fully-working system, they have an incentive to protect their own infrastructure as much as the United States does to continue to provide GPS and positioning more broadly to the highest standards of reliability.
Google is introducing new tools for newsrooms looking to understand their online audiences and how those audiences feed into their overall business.
These efforts fall under the umbrella of the broader Google News Initiative, introduced in 2018 as a way for the search giant to fund quality journalism and find other ways to support the industry. Since then, Google has introduced two journalism-focused products that sit on top of Google Analytics — News Consumer Insights, which is designed to help publishers grow their audiences and become more profitable, as well as Realtime Content Insights, which is supposed to help newsrooms see what’s trending at any given moment.
“We heard over and over again that our news partners were drowning in this data,” said Amy Adams Harding, director of analytics and revenue optimization for news and publishing at Google. “They had difficulty teasing out actionable intelligence in the tsunami of numbers.”
She added that it was “important” to her team that these products be free to use and “accessible to anyone using Google Analytics.”
Today, Google is introducing version 2.0 of both News Consumer Insights and Realtime Content Insights, while also adding a new feature called the News Tagging Guide.
Image Credits: Google
NTG is supposed to make it easier for publishers to collect the data they need. That falls into three broad categories — video analytics, user engagement and reader revenue. Publishers will be able to select the category and the specific types of data they want to track, and then Google will give them JavaScript that they can copy and paste onto their website in order to start feeding that data into Google Analytics.
Meanwhile, the News Consumer Insights product now includes personalized recommendations for the publisher. For example, it might point out that a publisher’s newsletter signups are relatively low, and then it could suggest different ways to improve those signups. Harding noted that NCI didn’t lack recommendations before, but to find them, publishers had to continually refer back to the generalized playbook, rather than having the most relevant recommendations highlighted for them while they’re going over their data.
And Realtime Content Insights have been expanded to include similar data about video content, as well as historic performance data, so publishers can see which stories performed best in a given period of time. And it’s not simply focused on pageviews — RCI also tracks things like social sharing and engagement, and it identifies which stories are doing better with casual readers versus loyal readers (who visit more than once a month) versus brand loyalists (who visit at least 15 times a month).
“We’re not saying one is better than the other,” said Anntao Diaz, Google’s head of News Consumer Insights, Realtime Content Insights and Google Surveys for Publishers. Instead, he suggested that different articles can attract different readers to serve different purposes, whether that’s growing the overall audience or building a relationship with a loyal audience that might pay for subscriptions.
Google has already been testing these features with select publishers, including Time and local newspaper publisher Lee Enterprises.
“As Lee Enterprises’ audience reach continues to grow in our newspapers’ communities, our partnership with Google News Initiative provides excellent insight and industry benchmarks to measure our success,” said Kyle Rickhoff, director of analytics at Lee Enterprises, in a statement. “The new version News Consumer Insights powered by News Tagging Guide improves our understanding of readers’ engagement and helps us prioritize business opportunities for video, engagement conversions and on-site subscriptions with better data.”
A year after its initial release, Jumbo has two important pieces of news to announce. First, the company has released a major update of its app that protects your privacy on online services. Second, the company has raised an $8 million Series A funding round.
If you’re not familiar with Jumbo, the app wants to fix what’s broken with online privacy today. Complicated terms of services combined customer-hostile default settings have made it really hard to understand what personal information is out there. Due to recent regulatory changes, it’s now possible to change privacy settings on many services.
While it’s possible, it doesn’t mean it’s easy. If you’ve tried to adjust your privacy settings on Facebook or LinkedIn, you know that it’s a convoluted process with a lot of sub-menus and non-descriptive text.
Similarly, social networks have been around for more than a decade. While you were comfortable sharing photos and public messages with a small group of friends ten years ago, you don’t necessarily want to leave this content accessible to hundreds or even thousands of “friends” today.
The result is an iPhone and Android app that puts you in charge of your privacy. It’s essentially a dashboard that lets you control your privacy on the web. You first connect the app to various online services and you can then control those services from Jumbo. Jumbo doesn’t limit itself to what you can do with APIs as it can mimic Javascript calls on web pages that are unaccessible to the APIs.
For instance, if you connect your Facebook account, you can remove your profile from advertising lists, delete past searches, change the visibility of posts you’re tagged in and more. On Google, you can delete your history across multiple services — web searches, Chrome history, YouTube searches, Google Map activities, location history, etc.
More fundamentally, Jumbo challenges the fact that everything should remain online forever. Conversations you had six months ago might not be relevant today, so why can’t you delete those conversations?
Jumbo lets you delete and archive old tweets, Messenger conversations and old Facebook posts. The app can regularly scan your accounts and delete everything that is older than a certain threshold — it can be a month, a year or whatever you want.
While your friends will no longer be able to see that content, Jumbo archives everything in a tab called Vault.
With today’s update, everything has been refined. The main tab has been redesigned to inform you of what Jumbo has been doing over the past week. The company now uses background notifications to perform some tasks even if you’re not launching the app every day.
The data breach monitoring has been improved. Jumbo now uses SpyCloud to tell you exactly what has been leaked in a data breach — your phone number, your email address, your password, your address, etc.
It’s also much easier to understand the settings you can change for each service thanks to simple toggles and recommendations that you can accept or ignore.
A clear business model
Jumbo’s basic features are free, but you’ll need to buy a subscription to access the most advanced features. Jumbo Plus lets you scan and archive your Instagram account, delete your Alexa voice recordings, manage your Reddit and Dropbox accounts, and track more than one email address for data breaches.
Jumbo Pro lets you manage your LinkedIn account (and you know that LinkedIn’s privacy settings are a mess). You can also track more information as part of the data breach feature — your ID, your credit card number and your social security number. It also lets you activate a tracker blocker.
This new feature in the second version of Jumbo replaces default DNS settings on your phone. All DNS requests are routed through a Jumbo-managed networking profile on your phone. If you’re trying to access a tracker, the request is blocked, if you’re trying to access some legit content, the request goes through. It works in the browser and in native apps.
You can pay what you want for Jumbo Plus, from $3 per month to $8 per month. Similarly, you can pick what you want to pay for Jumbo Pro between $9 per month and $15 per month.
You might think that you’re giving a ton of personal information to a small startup. Jumbo is well aware of that and tries to reassure its user base with radical design choices, transparency and a clear business model.
Jumbo doesn’t want to mine your data. Your archived data isn’t stored on Jumbo’s servers. It remains on your phone and optionally on your iCloud or Dropbox account as a backup.
Jumbo doesn’t even have user accounts. When you first open the app, the app assigns you with a unique ID in order to send you push notifications but that’s about it. The company has also hired companies for security audits.
“We don’t store email addresses so we don’t know why people subscribe,” Jumbo CEO Pierre Valade told me.
Profitable by 2022
Jumbo has raised an $8 million funding round. It had previously raised a $3.5 million seed round. This time, Balderton Capital is leading the round. The firm had already invested in Valade’s previous startup, Sunrise.
A lot of business angels participated in the round as well, and Jumbo is listing them all on its website. This is all about being transparent again.
Interestingly, Jumbo isn’t betting on explosive growth and eyeballs. The company says it has enough funding until February 2022. By then, the startup hopes it can attract 100,000 subscribers to reach profitability.
In a statement on Wednesday, the Competition Commission of India said it had approved Facebook’s proposed multi-billion-dollar investment in Jio Platforms for a 9.99% stake in the top Indian telecom network.
Jaadhu Holdings LLC, a wholly-owned subsidiary of Facebook, is acquiring the stake in Jio Platforms. Facebook created this subsidiary earlier this year.
The announcement comes a week after the watchdog said it was accessing the deal for potential misuse of users’ data and pondering if it should consider amending the current rules for some mergers and acquisitions in the country.
At the time, Facebook had argued that its investment in the Indian firm is “pro-competitive, benefits consumers, kirana stores (neighborhood stores) and other small and micro local Indian businesses, and take forward the vision of digital India.”
Analysts have said that Facebook’s investment in billionaire Mukesh Ambani’s Jio Platforms, its biggest investment in recent years, could help the social media giant expand its reach in India, which is already its biggest market by user count.
Facebook’s eponymous service reaches about 350 million users in India, while its messaging service WhatsApp has amassed over 400 million users. WhatsApp is by far the most popular service in the world’s second largest market.
In April, Facebook said it planned to work with Reliance Jio Platforms to empower 60 million small businesses, including mom-and-pop stores in India. Early signs of this collaboration was apparent a week later when JioMart, a joint venture between Reliance Jio Platforms and Reliance Retail (India’s largest retail chain), started to allow customers to track shipment through WhatsApp.
Some analysts said that the deal with Ambani, India’s richest man and an ally of Indian Prime Minister Narendra Modi, could also help Facebook stay on the good side of the Indian government. In India, where Facebook’s Free Basics program was blocked in early 2016, the firm has been stuck in a regulatory maze to get clearance for a nationwide rollout of WhatsApp Pay.
Spotify has launched a new playlist called Your Summer Rewind. Which, as the name suggests, is designed to transport you back to summers long gone. Essentially, Spotify knows what you listen to in summer, and has compiled these songs into a playlist.
Your Summer Rewind isn’t completely new. Spotify first launched Your Summer Rewind in June 2017, reminding listeners of previous summers. And now, in 2020, it’s back, with Spotify describing the playlist as “a new playlist featuring your old summer favorites.”
Your Summer Rewind is full of songs you have listened to on Spotify in previous summers. They may not be summery songs, but you happen to like listening to them in summer. Mine includes songs from bands I have seen in the summer and songs I play while camping.
You should find Your Summer Rewind on the homepage of the Spotify app. As with all of Spotify’s algorithmically produced playlists, it will be waiting for you in the “Made for [Your Name]” section. That is at least if you’re in the Northern Hemisphere.
Visit summers past, present, and future with our Songs of Summer 2020 predictions, Your Summer Rewind, and more.??https://t.co/xYaI1heMrs
As well as Your Summer Rewind, Spotify has revealed its 2020 Songs of Summer predictions. These are the songs the streaming service thinks will soundtrack summer 2020. Even though the ongoing COVID-19 pandemic may stifle many people’s plans.
Spotify’s predictions for summer 2020 include new songs such as ROCKSTAR by DaBaby, feat. Roddy Ricch and Savage Remix by Megan Thee Stallion, feat. Beyoncé. Plus old songs such as This Is America by Childish Gambino and Alright by Kendrick Lamar.
Given everything that’s been happening in the world of late, it’s easy to forget that we’re heading into summer in the Northern Hemisphere. However, Spotify’s Your Summer Rewind and Songs of Summer playlists should evoke memories old and new of this time of year.
On most physical keyboards, the Caps lock key size is relatively large. Because of its size and convenient position, we often inadvertently press the Caps lock key. We notice that the Caps lock is turned on only after typing a couple of letters, if not words. Luckily, Windows 10 offers a way to enable sound […]
The App Store ecosystem today is home to nearly 2 million apps. That means finding new apps to download is now more challenging than ever. This, in turn, leads app developers to funnel more money into App Store Search ads, traditional SEO and digital advertising in an effort to acquire new users. A new feature called App Clips, arriving in iOS 14 later this year, will give developers another option to introduce their app to users. With App Clips, users can instead load just a small part of an app on demand, when required. And when they’re done, the App Clip disappears.
The concept behind App Clips isn’t new. Google’s Android platform has for several years offered tiny app-on-demand downloads called “Instant Apps.”
Like Instant Apps, Apple’s App Clips are about making apps as seamless to use as the web. They are fast, ephemeral and eliminate the barrier to entry that is downloading an app from the App Store.
Today, many users don’t want to bother with a full app download when they’re in a hurry. For example, if a user is trying to pay for parking, they’re more likely to swipe their credit card in the meter to save time, instead of downloading the city’s parking app.
A customer waiting in line to place a food or drink order also doesn’t want to bother downloading the restaurant’s app to browse a menu and pay — they’ll just speak their order at the counter. And a customer wanting to rent a scooter just wants to tap, pay and be on their way.
Image Credits: Apple
An App Clip would work in any of of these scenarios, and many others, by making it as easy to use apps as it is to tap to check out with Apple Pay or launch a website.
While Apple will allow users to launch clips by way of a QR code, a new “App Clip Code” arriving later this year will offer an upgraded experience to kicking off these apps you find suggested to you out in the real world. App Clip Codes will combine both NFC and a visual code, so users can either tap or scan the code to access the App Clip experience.
Image Credits: Apple
For example, an App Clip Code placed on a parking meter would allow a user to quickly load just the part of the app where they pay for their time. They can even skip manual credit card entry by using Apple Pay, if included in a given App Clip.
The App Clips themselves are less than 10 MB in size and ship bundled with the app on the App Store. They’re built using the same UI technologies developers use today to build apps, like UIKit or SwiftUI. But using an App Clip doesn’t trigger the app to download to the user’s device.
A key advantage App Clips offer is how they address concerns over data privacy. Because App Clips are essentially a way to run app code on demand, they’re restricted from tapping into iPhone’s more sensitive data — like health and fitness information, for example. Plus, the App Clip and all its data will automatically disappear if it’s not used again within some period of time.
However, if a user begins to launch a particular App Clip more regularly — perhaps one for their favorite coffee order at their local shop, for instance — the App Clip’s lifetime is extended and it can get smarter. In this example, the App Clip could cache the customer’s last order and present it as a recommendation, to speed up the ordering process. Eventually, this repeat user may decide to download the full app.
In that case, the hand-off is seamless as well — iOS will automatically migrate the authorizations for things like Camera, Microphone and Bluetooth access, which the App Clip had already requested. Select data can also be migrated.
Image Credits: Apple
There are other ways for users to encounter App Clips besides out in the real world, though that may be a primary use case.
Apple says App Clips can be sent as links in iMessage, popped up as a suggestion when you’re browsing a mobile site in Safari, shown on a business’s details page in Apple Maps or may appear in Siri’s Nearby suggestions.
The idea is that wherever a user may be on their device — or out in the world — the App Clip can be there, too.
A vision for the future of Afghanistan | Ashraf Ghani
Offering a vision of Afghanistan that goes beyond what's often depicted in the media, President Ashraf Ghani shares his thoughts on peacemaking, the true cost of war, the nation's COVID-19 response strategy and the sweeping economic and social reforms happening throughout the country. "The ultimate goal is a sovereign, democratic, united Afghanistan at peace with itself and the world," he says. (This virtual conversation, hosted by head of TED Chris Anderson, was recorded June 16, 2020.)
Twitter took its latest action on content from President Trump Tuesday, again hiding a threat of state violence behind a warning label and appending it with a notice.
Trump’s latest offending tweet declared “There will never be an ‘Autonomous Zone’ in Washington, D.C., as long as I’m your President. If they try they will be met with serious force!” The tweet follows a clash between protesters and law enforcement Monday night in Lafayette Square near the White House.
We’ve placed a public interest notice on this Tweet for violating our policy against abusive behavior, specifically, the presence of a threat of harm against an identifiable group.https://t.co/AcmW6O6d4t
In recent weeks, the president has frequently derided the city of Seattle for allowing protesters to create a police-free area, returning to the topic to stoke fear and anger within his base. After police abandoned a station in the Capitol Hill neighborhood, Seattle protesters moved into the area, declaring it their own. That self-declared “autonomous zone”—and the president’s latest threat—grew out of national civil rights protests against police brutality and racist violence after Minneapolis police killed George Floyd late last month.
Twitter says the tweet violated its rules prohibiting threats of harm against groups of people, a category of “abusive behavior” on the social network. The company said that it will allow the tweet to remain up, though has restricted the ability of users to interact with it, including likes, replies and retweets without comment.
The tweet is the latest in what may become many examples of Twitter enforcing its platform rules against the president. In the past, the company rarely acted to enforce its policies on tweets from high profile U.S. politicians, Trump included.
Over the last month, Twitter shifted to a much more active approach to its moderation responsibilities for political figures. In late May, Twitter ignited a political firestorm when it flagged two of the president’s tweets making false claims about vote-by-mail systems in California, leading to a retaliatory executive order from Trump days later. In the early days of the George Floyd protests, the company hid another tweet from the president that threatened lethal violence against protesters.
Two months after Quibi’s high-profile launch as a short-form mobile-native TV app led by Jeffrey Katzenberg and Meg Whitman, it is evident the startup is greatly underperforming relative to the hundreds of millions of dollars already spent on content and marketing.
According to aWall Street Journal report, “daily downloads peaked at 379,000 on its April 6 launch day but didn’t exceed 20,000 on any day in the first week of June, according to Sensor Tower.” The article says Quibi is on pace for just 2 million subscribers by year-end, from its predicted 7.2 million. Most of the current subscriber base is on free trials, so even just maintaining the current pace of subscriber growth for several more months will be challenging. Quibi hasn’t released any of its own stats on subscribers, which it almost certainly would do to combat the negative perception among investors and press, if the stats showed a lot of traction.
Iargued in 2018that Facebook should turn its IGTV into a Quibi competitor, and I continue to believe there’s untapped opportunity for premium, mobile-native storytelling apps. So what went wrong with Quibi? There appear to have been four key mistakes:
Miscalculating the risk of launching during the COVID-19 lockdown.
Failing to see the central role of interactivity in mobile-native entertainment.
Creating misaligned financial incentives with the wrong content partners.
Launching Quibi like a movie instead of like a startup.
Snap’s latest developer offering is a tool called Verify that’s aiming to save its mobile-first developers cash on verifying user phone numbers en masse. The new tool is being integrated into its Login Kit framework, allowing devs with a “Log In with Snapchat” button to move away from tapping services like Twilio for SMS verification, instead checking with Snap to see whether that user has already verified their number with Snapchat.
Snap Kit lead Patrick Mandia tells TechCrunch that his team has been working on the feature for more than a year, aiming to create “an easy, private way to verify phone numbers at no cost to developers.” Mandia says that because phone number verification is already a core part of Snapchat’s login flow, the company wanted to create a way for companies inside its SDK to verify phone numbers with them rather than tapping an external service.
Verify basically works by checking whether the phone number a user just typed into the login flow of a Snap Kit app matches the phone number associated with the Snapchat account they used to log in to the app.
Rather than sending users a verification code, Snap Kit handles this all in the background. If the user has verified that number within the past year, Snap will okay the verification. If a user doesn’t have a phone number on file with Snapchat or hasn’t recently verified that number, Snap will carry out that SMS verification on their end with the user, keeping it on file so the number can be confirmed through Verify on other Snap Kit applications.
At its annual partner summit last week, Snap shared that it now has 800 developers on its Snap Kit platform that collectively reach 150 million users on a monthly basis. The social messaging company has been looking to grow that network through features that tap into its network of resources rather than its network of user data.
“Traditionally, when you think of these platforms, the value is data. The value is what data a third party is going to be able to get about you and your friends, and that turns out to have some real issues,” Snap’s VP of Partnerships Ben Schwerin told us. “So, we took a much different approach, which is that Login Kit is going to power really creative, unique experiences and that’s going to be the value for partners.”
In a conversation with Snap last week, the company connected me with Yolo co-founder Greg Henrion whose app operates on the Snap Kit platform. He says that his app is spending “above six-figures” on a monthly basis just to verify phone numbers.
“The idea that we have portfolio companies spending $1 million per year to verify text messages is pretty crazy,” says Josh Miller, an investor at Thrive Capital, which backed Yolo.
For consumer app developers with a user base that overlaps significantly with Snapchat’s, the new feature offers them something that could save them cash right off the bat. Yolo’s situation is certainly unique, Snap Kit is the only way to log in to their platform, so Verify will take care of almost all of their SMS verification spend. Yolo’s outsized spending on SMS verification is also an outlier; the startup says about one-third of their run rate is devoted to it.
Verify with Snapchat is free and is available for Snap Kit developers starting today.
Posted by Eunsol Choi, Visiting Faculty Researcher and Tom Kwiatkowski, Research Scientist, Google Research
One of the primary goals of natural language processing is to build systems that can answer a user's questions. To do this, computers need to be able to understand questions, represent world knowledge, and reason their way to answers. Traditionally, answers have been retrieved from a collection of documents or a knowledge graph. For example, to answer the question, “When was the declaration of independence officially signed?” a system might first find the most relevant article from Wikipedia, and then locate a sentence containing the answer, “August 2, 1776”. However, more recent approaches, like T5, have also shown that neural models, trained on large amounts of web-text, can also answer questions directly, without retrieving documents or facts from a knowledge graph. This has led to significant debate about how knowledge should be stored for use by our question answering systems — in human readable text and structured formats, or in the learned parameters of a neural network.
Today, we are proud to announce the EfficientQA competition and workshop at NeurIPS 2020, organized in cooperation with Princeton University and the University of Washington. The goal is to develop an end-to-end question answering system that contains all of the knowledge required to answer open-domain questions. There are no constraints on how the knowledge is stored — it could be in documents, databases, the parameters of a neural network, or any other form — but entries will be evaluated based on the number of bytes used to access this knowledge, including code, corpora, and model parameters. There will also be an unconstrained track, in which the goal is to achieve the best possible question answering performance regardless of system size. To build small, yet robust systems, participants will have to explore new methods of knowledge representation and reasoning.
An illustration of how the memory budget changes as a neural network and retrieval corpus grow and shrink. It is possible that successful systems will also use other resources such as a knowledge graph.
Competition Overview
The competition will be evaluated using the open-domain variant of the Natural Questions dataset. We will also provide further human evaluation of all the top performing entries to account for the fact that there are many correct ways to answer a question, not all of which will be covered by any set of reference answers. For example, for the question “What type of car is a Jeep considered?” both “off-road vehicles” and “crossover SUVs” are valid answers.
The competition is divided between four separate tracks: best performing system under 500 Mb; best performing system under 6 Gb; smallest system to get at least 25% accuracy; and the best performing system with no constraints. The winners of each of these tracks will be invited to present their work during the competition track at NeurIPS 2020, which will be hosted virtually. We will also put each of the winning systems up against human trivia experts (the 2017 NeurIPS Human-Computer competition featured Jeopardy! and Who Wants to Be a Millionaire champions) in a real-time contest at the virtual conference.
Participation
To participate, go to the competition site where you will find the data and evaluation code available for download, as well as dates and instructions on how to participate, and a sign-up form for updates. Along with our academic collaborators, we have provided some example systems to help you get started.
We believe that the field of natural language processing will benefit from a greater exploration and comparison of small system question answering options. We hope that by encouraging the development of very small systems, this competition will pave the way for on-device question answering.
Acknowledgements
Creating this challenge and workshop has been a large team effort including Adam Roberts, Colin Raffel, Chris Alberti, Jordan Boyd-Graber, Jennimaria Palomaki, Kenton Lee, Kelvin Guu, and Michael Collins from Google; as well as Sewon Min and Hannaneh Hajishirzi from the University of Washington; and Danqi Chen from Princeton University.
