AltSolution

Thank you, Chrome team 🤩🤩🤩

Since Chrome came out back in 2008, it’s been a constant companion in my life. In fact, Chrome’s launch is how I helped get the startup I worked for at the time onto TechCrunch for the first time.

We did shots to celebrate. Chrome rocked, and we were Day One Fans.

But over time what was once a romance began to sour, as Chrome got a bit slower, a bit heavier and a bit worse over the years.

The devolution felt a bit like what was happening to Google search, in which a very good idea was slowly turned into something that made more money at the cost of functionality, speed and user happiness (more on that natural terminus of that progression here).

And because I am a petulant child, I have been very annoyed by what has happened to Chrome, software that I have never paid a single dollar to use. To make this point, I went out to round up a tweet or two from myself complaining about Chrome over the years, but after finding at least nine examples since May I started to feel bad (one, two, three, four, five, six, seven, eight, nine). So let’s move on.

What went wrong with Chrome? I don’t know. Over time its taste for RAM, lag and being generally annoying grew. But as I was living in a G Suite world, sticking to Chrome made sense — so I endured.

And now, I may not have to any longer. This week Google detailed an impending set of Chrome updates that are amazing to read through and imagine the real-world impact of. Big Goog appears to have gone deep into its browser’s code, finding ways to make it faster, lighter on memory usage and smarter.

I am so very excited.

What’s coming? Pulling from Google’s Chromium blog instead of its more consumer-friendly post (a big thanks to The Verge for bringing this set of updates to my attention), here are the highlights as far as I am concerned (Bolding: TechCrunch in each block quote):

Even if you have a lot of tabs open, you likely only focus on a small set of them to get a task done. Starting in this release, Chrome is actively managing your computer’s resources to make the tabs you care about fast—while allowing you to keep hundreds of tabs open—so you can pick up where you left off.

In this release, we’re improving how Chrome understands and manages resources with Tab throttling, occlusion tracking and back/forward caching, so you can quickly get to what you need when you need it.

Google this is literally me. I feel incredibly seen. Thank you.

We investigated how background tabs use system resources and found that JavaScript Timers represent >40% of the work in background tabs. Reducing their impact on CPU and power is important to make the browser more efficient. Beginning in M87, we’re throttling JavaScript timer wake-ups in background tabs to once per minute. This reduces CPU usage by up to 5x, and extends battery life up to 1.25 hours in our internal testing.

When the world works again, I want to buy lunch for everyone who took part in this effort.

Next, we’re bringing Occlusion Tracking–which was previously added to Chrome OS and Mac–to Windows, which allows Chrome to know which windows and tabs are actually visible to you. With this information, Chrome can optimize resources for the tabs you are using, not the ones you’ve minimized, making Chrome up to 25% faster to start up and 7% faster to load pages, all while using less memory.

Hell yes.

How many times have you visited a website and clicked a link to go to another page, only to realize it’s not what you wanted and click the back button? […] In Chrome 87, our back/forward cache will make 20% of those back/forward navigations instant, with plans to increase this to 50% through further improvements and developer outreach in the near future.

I didn’t even know I needed this, but I do. And I can’t wait to have it.

All in all, as I write this short post to you inside of Chrome, I cannot help but be freaking excited about New And Improved Chrome. More later after I get some testing in, but, honestly, yay!
https://tcrn.ch/2IZ497m

AI researchers made a sarcasm detection model and it’s sooo impressive 😮🤨😏

Researchers in China say they’ve created sarcasm detection AI that achieved state-of-the-art performance on a dataset drawn from Twitter. The AI uses multimodal learning that combines text and imagery since both are often needed to understand whether a person is being sarcastic.

The researchers argue that sarcasm detection can assist with sentiment analysis and crowdsourced understanding of public attitudes about a particular subject. In a challenge initiated earlier this year, Facebook is using multimodal AI to recognize whether memes violate its terms of service.

The researchers’ AI focuses on differences between text and imagery and then combines those results to make predictions. It also compares hashtags to tweet text to help assess the sentiment a user is trying to convey.

“Particularly, the input tokens will give high attention values to the image regions contradicting them, as incongruity is a key character of sarcasm,” the paper reads. “As the incongruity might only appear within the text (e.g., a sarcastic text associated with an unrelated image), it is necessary to consider the intra modality incongruity.”

On a dataset drawn from Twitter, the model achieved a 2.74% improvement on a sarcasm detection F1 score compared to HFM, a multimodal detection model introduced last year. The new model also achieved an 86% accuracy rate, compared to 83% for HFM.

The paper was published jointly by the Chinese Academy of Sciences and the Institute of Information Engineering, both in Beijing, China. The paper was presented this week at the virtual Empirical Methods in Natural Language Processing (EMNLP) conference.

The AI is the latest example of multimodal sarcasm detection to emerge since AI researchers began studying sarcasm in multimodal content on Instagram, Tumblr, and Twitter in 2016.

University of Michigan and University of Singapore researchers used language models and computer vision to detect sarcasm in television shows, a model detailed in a paper titled “Towards Multimodal Sarcasm Detection (An Obviously Perfect Paper).” That work was highlighted as part of the Association for Computational Linguistics (ACL) last year.
https://bit.ly/2IIKuJs

Google Maps gains driving mode and food delivery status 🤔

Google today announced updates to Google Maps aimed at informing users about risks related to the pandemic. Maps will soon show all-time detected COVID-19 cases in an area, along with quick links to resources from local authorities. Google will also begin to show how crowded bus, train, and subway lines are in more places around the world. In supported countries, Maps also now displays the live status of takeout and delivery orders, as well as reservations. And in the U.S., Maps features a new driving mode that puts Google Assistant front and center.

Google says it has added nearly 250 features and improvements to Maps since the start of the pandemic, including live busyness information for millions of places. Over 50 million updates are made to the map each day, drawing on data from more than 10,000 local governments, transit agencies, and organizations, according to the company. These include “popular times” information for over 20 million places worldwide.

Beyond the enhanced COVID-19 map layer on Android and iOS and the public transit crowdedness data, Maps users in the U.S., Canada, Germany, Australia, Brazil, and India will soon see food order information, including when to pick up orders, when to expect deliveries to arrive, anticipated wait times and delivery fees, and reorder shortcuts. Maps will also show reservation status in 70 countries around the world.

In addition, Maps is gaining an interface specifically optimized for hands-free driving. After previewing this new Google Assistant driving mode in Maps in early 2020, Google is today offering an early preview of the experience on Android in English in the U.S. With Google Assistant driving mode in Maps, users can leverage voice to send and receive calls and texts, review new messages across apps, and get a read-out of texts. Assistant will alert them to incoming calls so they can answer or decline with their voice. It will also respond to media playback commands for “hundreds” of providers, including YouTube Music, Spotify, Google Podcasts, and more.

To get started with Google Assistant driving mode, begin navigating to a destination with Maps and tap on the popup. Alternatively, head to Assistant settings on an Android phone or say “Hey Google, open Assistant settings,” then select “Getting around,” choose “Driving mode,” and switch it on.

“Driving mode makes all of this possible without ever leaving the navigation screen, so you can minimize distractions on the road,” Maps VP Dane Glasgow wrote in a blog post. “To make sure that information is as accurate and up-to-date as possible, we rely on 170 billion high-definition Street View images from 87 countries, contributions from hundreds of millions of businesses and people using Google Maps … We also invest in technical approaches that power some of our most beloved and essential features — from the 20 million places globally that now show popular times data to AR-powered Live View.”
https://bit.ly/396oFyj

Gmail users get new controls for data used to personalize Google’s ‘smart’ features 🙂🙂🙂

Google is introducing new controls for data it uses to personalize various “smart” features across its suite of products. This means Gmail users will soon be able to access a setting that stipulates whether their Gmail, Chat, or Meet data can be leveraged for Google’s automated Smart Reply or Smart Compose, for example, or whether bill payment reminders gleaned from email data can be issued through Google Assistant.

The launch comes as Google and other internet giants face increasing scrutiny over their data privacy practices, ushering in new regulations designed to protect consumers’ online privacy.

Smart data
Google has introduced myriad “smart” features that apply AI and machine learning techniques to user data to enhance its services. For example, Gmail users are likely familiar with Smart Compose, which is basically auto-complete for emails — it can use historical grammar and typing habits to make word and sentence suggestions in real time.

Amid concerns over the way Big Tech companies manipulate user data for their own purposes, Google has offered a few data privacy controls over the years, including dedicated security dashboards and privacy checkup tools that allow users to manage and deactivate such features as location history, web and app activity, and YouTube viewing history. Earlier this year, Google also activated a new auto-delete feature by default for some users.

In Gmail, users can already delve into their account settings and stipulate not only whether Smart Compose is activated, but whether it’s personalized to their writing style.

But even if users know these features exist, it’s easy to lose track of which settings are on and off.

Rolling out “in the coming weeks,” Google’s new setting aims to add centralized controls for Gmail data used to power its various smart features. Users who visit their account settings will now see an option asking whether they want to activate or deactivate smart features in Gmail, Chat, or Meet.

A second screen asks the user whether they also want to allow their Gmail, Chat, and Meet data to be used to personalize other Google services, such as Google Assistant, Google Maps, and Google Travel.

Along with features like bill reminders in Google Assistant, this will impact things like whether Google can create automated travel itineraries from trip bookings it detects in users’ Gmail messages.

Data play
Numerous data privacy laws have taken effect in the past few years, including GDPR in Europe and the CCPA in California. Companies like Google have had to change how they operate or face the wrath of legislators. Indeed, Google is currently facing complaints on numerous fronts, including an accusation by browser rival Brave that Google is breaching European data protection law by sharing data between its services without proper consent.

Closer to home, Google is facing a $5 billion class action lawsuit in California that alleges the internet giant essentially ignores users’ privacy wishes. And in the U.K., Google is facing another lawsuit over how YouTube uses kids’ data to target advertising.

Put simply, Google is fighting fires on many fronts.

The company said its new setting is less about adding privacy functionality than it is about giving users a “clearer choice over the data processing” that makes its smart features possible. Google said the feature was designed based on learnings from “user experience research and regulators’ emphasis on comprehensible, actionable user choices over data.”
https://bit.ly/3pvFJ6x

You can’t eliminate bias from machine learning, but you can pick your bias 🤨

Bias is a major topic of concern in mainstream society, which has embraced the concept that certain characteristics — race, gender, age, or zip code, for example — should not matter when making decisions about things such as credit or insurance. But while an absence of bias makes sense on a human level, in the world of machine learning, it’s a bit different.

In machine learning theory, if you can mathematically prove you don’t have any bias and if you find the optimal model, the value of the model actually diminishes because you will not be able to make generalizations. What this tells us is that, as unfortunate as it may sound, without any bias built into the model, you cannot learn.

The oxymoron of discrimination-free discriminators

Modern businesses want to use machine learning and data mining to make decisions based on what their data tells them, but the very nature of that inquiry is discriminatory. Yet, it is perhaps not discriminatory in the way that we typically define the word. The purpose of data mining is to, as Merriam-Webster puts it, “distinguish by discerning or exposing differences: to recognize or identify as separate and distinct,” rather than “to make a difference in treatment or favor on a basis other than individual merit.” It is a subtle but important distinction.

Society clearly passes judgments on people and treats them differently based on many different categories. Well-intentioned organizations try to rectify or overcompensate for this by eliminating bias in machine learning models. What they don’t realize is that in doing so, it can mess things up further. Why is this? Once you get into removing data categories, other components, characteristics, or traits sneak in.

Suppose, for example, you uncover that income is biasing your model, but there is also a correlation between income and where someone comes from (wages vary by geography). The moment you add income into the model, you need to de-discriminate that by putting origin in as well. It’s extremely hard to make sure that you have nothing discriminatory in the model. If you take out where someone comes from, how much they earn, where they live, and maybe what their education is, there’s not much left to allow you to determine the difference between one person to another. And still, there could be some remaining bias you haven’t thought about.

David Hand has described how the United Kingdom once mandated that car insurance policies couldn’t discriminate against young or old drivers, nor could they set different premiums by gender. On the surface, this sounds nice, how very equal. The problem is that people within these groupings generally have different accident rates. When age and gender are included in the data model, it shows young males have much higher accident rates, and the accidents are more serious; therefore, they should theoretically pay higher premiums.

By removing the gender and age categories, however, policy rates go down for young men, enabling more to afford insurance. In the UK model, this factor — more young men with insurance — ultimately drove up the number of overall accidents. The changed model also introduced a new type of bias: Women were paying a disproportionate amount for insurance compared to their accident ratio because they were sponsoring the increased number of accidents by young males. The example shows that you sometimes get undesired side effects by removing categories from the model. The moment you take something out, you haven’t necessarily eliminated bias. It’s still present in the data, only in a different way. When you get rid of a category, you start messing with the whole system.

We find a reverse of the above example in Germany. There, health insurers are not allowed to charge differently based on gender, even though men and women clearly experience different conditions and risk factors throughout their lives. For example, women generate significant costs to the health system around pregnancy and giving birth, but no one argues about it because the outcome is viewed as positive — vs. the negative association with car accidents in the UK — therefore, it is perceived as fair that those costs are distributed evenly.

The danger of omission
The omission of data is quite common, and it doesn’t just occur when you remove a category.

Suppose you’re trying to decide who is qualified for a loan. Even the best models will have a certain margin of error because you’re not looking at all of the people that didn’t end up getting a loan. Some people who wanted loans may have never come into the bank in the first place, or maybe they walked in and didn’t make it to your desk; they were scared away based on the environment or got nervous that they would not be successful.

As such, your model may not contain the comprehensive set of data points it needs to make a decision...
https://bit.ly/2H4ACZw

How Google built its AI-powered Hum to Search feature 🤔

In October, Google announced it would let users search for songs by simply humming or whistling melodies, initially in English on iOS and in more than 20 languages on Android. At the time, the search giant only hinted at how the new Hum to Search feature worked. But in a blog post today, Google detailed the underlying systems that enable Google Search to find songs using only hummed renditions.

Identifying songs from humming is a longstanding challenge in AI. With lyrics, background vocals, and a range of instruments, the audio of a musical or studio recording can be quite different from a hummed version. When someone hums their interpretation of a song, the pitch, key, tempo, and rhythm often vary slightly or significantly from the original. That’s why so many existing approaches to query by humming match the hummed tune against a database of preexisting hummed or melody-only versions of a song instead of identifying the song directly.

By contrast, Google’s Hum by Search matches a hummed melody directly to the original recordings without relying on a database of recordings paired with hummed versions of each. Google notes that this approach allows Hum to Search to be refreshed with millions of original recordings from across the world, including the latest releases.

This is just one example of how Google is applying AI to improve the Search experience. A recent algorithmic enhancement to Google’s spellchecker feature enabled more accurate and precise spelling suggestions. Search now leverages AI to capture the nuances of the webpage content it indexes. And Google says it is using computer vision to highlight notable points in videos within Search, like a screenshot comparing different products or a key step in a recipe.

Matching melodies
Hum to Search builds on Google’s extensive work in music recognition. In 2017, the company launched Now Playing with its Pixel smartphone lineup, which uses an on-device, offline machine learning algorithm and a database of song fingerprints to recognize music playing nearby. As it identifies a song, Now Playing records the track name in an on-device history. And if a Pixel is idle and charging while connected to Wi-Fi, a Google server sometimes invites it to join a “round” of computation with hundreds of other Pixel phones. The result enables Google engineers to improve the Now Playing song database without any phone revealing which songs were heard.

Google refined this technology in Sound Search, which provides a server-based recognition service to let users more quickly and accurately find over 100 million songs. Sound Search was built before the widespread use of machine learning algorithms, but Google revamped it in 2018 using scaled-up versions of the AI models powering Now Playing. Google also began weighing Sound Search’s index based on popularity, lowering the threshold for popular songs and raising it for obscure songs.

But matching hummed tunes with songs required a novel approach. As Google explains, it had to develop a model that could learn to focus on the dominant melody of a song while ignoring vocals, instruments, and voice timbre; differences stemming from background noises; and room reverberations.

A humming model
For Hum to Search, Google modified the music recognition models leveraged in Now Playing and Sound Search to work with hummed recordings. Google trained these retrieval models using pairs of hummed or sung audio with recorded audio to produce embeddings (i.e., numerical representations) for each input. In practice, the modified models produce embeddings with pairs of audio containing the same melody close to each other (even if they have different instrumental accompaniment and singing voices) and pairs of audio containing different melodies far apart. Finding the matching song is a matter of searching for similar embeddings from Google’s database of recordings.

Because training the models required song pairs — recorded songs and sung songs — the first barrier was obtaining enough training data. Google says its initial dataset consisted of mostly sung music segments (very few of which contained humming) and that it made the models more robust by augmenting the audio during training. It did this by varying the pitch or tempo of the sung input randomly, for example.

The resulting models worked well enough for people singing, but not for those humming or whistling. To rectify this, Google generated additional training data by simulating “hummed” melodies from the existing audio dataset using SPICE, a pitch extraction model developed by the company’s wider team as part of the FreddieMeter project. FreddieMeter uses on-device machine learning models developed by Google to see how close a person’s vocal timbre, pitch, and melody are to the artist Freddie Mercury...
https://bit.ly/38FP5GK

Google Photos will end free, unlimited storage 😮🤔

Google changes its storage policy, Facebook extends its political ad ban and Ring doorbells are recalled. This is your Daily Crunch for November 11, 2020.

The big story: Google Photos will end free, unlimited storage

Google is changing its storage policies for free accounts in a way that could have a big impact on anyone regularly using Google Photos.

Currently, Google Photos allows users to store unlimited images (and HD video) as long as they’re under 16 megapixels. Starting on June 1, 2021, new photos and videos will all count toward the 15 gigabytes of free storage that the company offers to anyone with a free Google account.

Google says it will take the average user three years to reach 15 gigabytes — at which point they’ll either need to delete some photos or pay for a Google One account. Also on June 1: Docs, Sheets, Slides, Drawings, Forms and Jamboard files will start counting toward your storage total as well.

The tech giants

Facebook extends its temporary ban on political ads for another month — The company says the temporary ban will continue for at least another month.

ByteDance asks federal appeals court to vacate US order forcing it to sell TikTok — TikTok’s parent company says it remains committed to a negotiated solution and will only try to stop the government from forcing a sale “if discussions reach an impasse.”

Ring doorbells recalled over fire threat — The recall comes in the wake of 23 reports of fire and eight reports of minor burns.

Startups, funding and venture capital

SentinelOne, an AI-based endpoint security firm, confirms $267M raise on a $3.1B valuation — SentinelOne’s Singularity monitors and secures laptops, phones and other network-connected devices and services.

E-commerce startup Heroes raises $65M in equity and debt to become the Thrasio of Europe — The company has a strategy of acquiring and scaling high-performing Amazon businesses.

Seedcamp raises £78M for its fifth fund — This new fund increases the amount of capital the firm will invest in pre-seed and seed-stage companies.

Advice and analysis from Extra Crunch

Dear Sophie: What does Biden’s win mean for tech immigration? — Attorney Sophie Alcorn looks at the presidential election’s impact on U.S. immigration and immigration reform.

Greylock’s Asheem Chandna on ‘shifting left’ in cybersecurity and the future of enterprise startups — Enterprise software is changing faster this year than it has in a decade.

Square and PayPal earnings bring good (and bad) news for fintech startups — Square’s earnings give us a window into consumer payment activity, card usage, stock purchases and more.

(Reminder: Extra Crunch is our membership program, which aims to democratize information about startups. You can sign up here.)

Everything else

Honda to mass-produce Level 3 autonomous cars by March — Honda claims it will be the first automaker to mass-produce vehicles with autonomous capabilities that meet SAE Level 3 standards.

Data audit of UK political parties finds laundry list of failings — The audit claims parties are failing to come clean with voters about how they’re being invisibly profiled and targeted.
https://tcrn.ch/2IkCIoY

Google launches Document AI suite of parsing and processing tools in preview 🤔

Google this morning launched the Document AI (DocAI) platform, a console for document processing hosted in Google Cloud, in preview. The company says it’s aimed at automating and validating documents by extracting data from documents and making them available to business apps and users.

Companies spend an average of $20 to file and store a single document, by some estimates, and only 18% of companies consider themselves paperless. An IDC report revealed that document-related challenges account for a 21.3% productivity loss, and U.S. companies waste a collective $8 billion annually managing paperwork.

Google’s DocAI platform ostensibly solves this by providing access to document parsers, tools, and solutions via an API. It supports the creation and customization of document processing workflows built with Google Cloud’s predefined taxonomy without the need to perform additional data mapping or training. DocAI offers general processors including a form parser, W9 parser, optical character recognition, document splitter, and custom workflows for domain-specific documents. These reside in a unified dashboard from where they can be tested by uploading a document directly in the console.

The parsers can classify information in documents like addresses, account numbers, and signatures as well as extract data like supplier names, invoice dates, and payment terms. Google says it’s working on additional capabilities for the DocAI platform to grow its core capabilities and support additional toolsets.

General parsers such as optical character recognition, the form parser, and the document splitter are available from the DocAI platform console. Access to specialized parsers like W9, 1040, W2, 1099-MISC, 1003, invoice, and receipts must be requested on a per-customer basis.

The launch of Google’s DocAI platform comes after the release of Lending DocAI, a Google Cloud product for the mortgage industry that ostensibly provides “industry-leading” accuracy for documents relevant to lending and processing. Google also recently unveiled PPP Lending AI, an effort to help lenders expedite the processing of applications for the since-exhausted U.S. Small Business Administration’s (SBA) Paycheck Protection Program, and Procurement DocAI, which automates procurement data capture by turning docs like invoices and receipts into structured data.

Lending DocAI and Procurement DocAI are now a part of the DocAI platform.

“We believe that any company that has to manually extract data from complex documents at scale can greatly benefit from Google Cloud AI,” Google product manager Lewis Liu and product marketing manager Yang Liang wrote in a blog post. “Transforming documents into structured data increases the speed of decision making for companies, unlocking measurable business value and helping develop better experiences for customers.”
https://bit.ly/3ndbS0q