Year: 2020

YouTube today announced it’s launching a new feature that will push commenters to reconsider their hateful and offensive remarks before posting. It will also begin testing a filter that allows creators to avoid having to read some of the hurtful comments on their channel that had been automatically held for review. The new features are meant to address long standing issues with the quality of comments on YouTube’s platform — a problem creators have complained about for years.

The company said it will also soon run a survey aimed at giving equal opportunity to creators, and whose data can help the company to better understand how some creators are more disproportionately impacted by online hate and harassment.

The new commenting feature, rolling out today, is a significant change for YouTube.

The feature appears when users are about to post something offensive in a video’s comments section and warns to “Keep comments respectful.” The message also tells users to check the site’s Community Guidelines if they’re not sure if a comment is appropriate.

The pop-up then nudges users to click the “Edit” button and revise their comment by making “Edit” the more prominent choice on the screen that appears.

The feature will not actually prevent a user from posting their comment, however. If they want to proceed, they can click the “Post Anyway” option instead.

Image Credits: YouTube

The idea to put up roadblocks to give users time to pause and reconsider their words and actions is something several social media platforms are now doing.

For instance, Instagram last year launched a feature that would flag offensive comments before they were posted. It later expanded that to include offensive captions. Without providing data, the company claimed that these “nudges” were helping to reduce online bullying. Meanwhile, Twitter this year began to push users to read the article linked in tweets they were about to share before tweeting their reaction, and it stopped users from being able to retweet with just one click.

These intentional pauses built into the social platforms are designed to stop people from reacting to content with heightened emotion and anger, and instead push users to be more thoughtful in what they say and do. User interface changes like this leverage basic human psychology to work, and may even prove effective in some percentage of cases. But platforms have been hesitant to roll out such tweaks as they can stifle user engagement.

In YouTube’s case, the company tells TechCrunch its systems will learn what’s considered offensive based on what content gets repeatedly flagged by users. Over time, this A.I.-powered system should be able to improve as the technology gets better at detection and the system itself is further developed.

Users on Android in the English language will see the new prompts first, starting today, Google says. The rollout will complete over the next couple of days. The company did not offer a timeframe for the feature’s support for other platforms and languages or even a firm commitment that such support would arrive in the future.

In addition, YouTube said it will also now begin testing a feature for creators who use YouTube Studio to manage their channel.

Creators will be able to try out a new filter that will hide the offensive and hurtful comments that have automatically been held for review.

Today, YouTube Studio users can choose to auto-moderate potentially inappropriate comments, which they can then manually review and choose to approve, hide or report. While it’s helpful to have these held, it’s still often difficult for creators to have to deal with these comments at all, as online trolls can be unbelievably cruel. With the filter, creators can avoid these potentially offensive comments entirely.

YouTube says it will also streamline its moderation tools to make the review process easier going forward.

The changes follow a year during which YouTube has been heavily criticized for not doing enough to combat hate speech and misinformation on its platform. The video platform’s “strikes” system for rule violations means that videos may be individually removed but a channel itself can stay online unless it collects enough strikes to be taken down. In practice, that means a YouTube creator could be as violent as calling for government officials to be beheaded and and still continue to use YouTube. (By comparison, that same threat led to an account ban on Twitter.)

YouTube claims it has increased the number of daily hate speech comment removals by 46x since early 2019. And in the last quarter, of the more than 1.8 million channels it terminated for violating our policies, more than 54,000 terminations were for hate speech. That indicates a growing problem with online discourse that likely influenced these new measures. Some would argue the platforms have a responsibility to do even more, but it’s a difficult balance.

In a separate move, YouTube said it’s soon introducing a new survey that will ask creators to voluntarily share with YouTube information about their gender, sexual orientation, race and ethnicity. Using the data collected, YouTube claims it will be able to better examine how content from different communities is treated in search, discovery and monetization systems.

It will also look for possible patterns of hate, harassment, and discrimination that could affect some communities more than others, the company explains. And the survey will give creators to optionally participate in other initiatives that YouTube hosts, like #YouTubeBlack creator gatherings or FanFest, for instance.

This survey will begin in 2021 and was designed in consultation with input from creators and civil and human rights experts. YouTube says the collected data will not be used for advertising purposes, and creators will have the ability to opt-out and delete their information entirely at any time.

Timnit Gebru, a leading researcher and voice in the field of ethics and artificial intelligence, says Google fired her for an email she sent to her direct reports.  According to Gebru, Google fired her because of an email she sent to subordinates that the company said reflected “behavior that is inconsistent with the expectations of a Google manager.”

Gebru, the co-leader of Google Ethical Artificial Intelligence team, took to Twitter last night, shortly after the National Labor Relations Board filed a complaint against Google alleging surveillance of employees and unlawful firing of employees.

Gebru says no one explicitly told her she was fired, but that she received an email from one of her boss’s reports, saying:

“Thanks for making your conditions clear. We cannot agree to #1 and #2 as you are requesting. We respect your decision to leave Google as a result, and we are accepting your resignation.”

That email, according to Gebru, went on to say that “certain aspects of the email you sent last night to non-management employees in the brain group reflect behavior that is inconsistent with the expectations of a Google manager.”

It’s not clear what exactly was contained in the email. We’ve reached out to both Gebru and Google for comment.

As Bloomberg reported, Gebru has been outspoken about the lack of diversity in tech as well as the injustices Black people in tech face. According to Bloomberg, Gebru believes Google let her go to signal to other workers that it’s not ok to speak up.

Gebru is a leading voice in the field of ethics and artificial intelligence. In 2018, Gebru collaborated with Joy Buolamwini, founder of the Algorithmic Justice League, to study biases in facial recognition systems. They found high disparities in error rates between lighter males and darker females, which led to the conclusion that those systems didn’t work well for people with darker skin.

Since Gebru’s announcement, she’s received an outpouring of support from those in the tech community.

Developing…

General Catalyst is one of the top VC firms in the U.S., with portfolio companies that include Snap, Kayak, Airbnb, Stripe, HubSpot, GitLab and many others.

We’re thrilled to have GC’s Peter Boyce and Katherine Boyle join us for the next episode of Extra Crunch Live, our live video series where we ask VCs about what’s exciting them these days, get their advice for how early-stage startups can thrive (particularly during a pandemic), and offer Extra Crunch members a chance to ask questions directly.

We’ve interviewed some heavy hitters, including notables like Mark Cuban, Kirsten Green and Roelof Botha during our first two seasons.

If you’re not an Extra Crunch member, you should really hop to it; check out the full library of ECL content before hanging out with Boyce, Boyle and myself on Tuesday at 4pm ET/1pm PT.

Boyce has been with General Catalyst since 2013, leading investments in companies such as Ro, Macro, towerIQ and Atom, among others. He also supported some big General Catalyst deals, including investments in Giphy, Jet.com and Circle.

Boyce also co-founded Rough Draft Ventures, an investment arm of General Catalyst focused on funding first-time CEOs out of university.

Boyle was previously a business reporter at The Washington Post before joining General Catalyst, which gives her a unique perspective on the entrepreneurial landscape. She’s invested in several companies, including AirMap, Origin and Nova Credit.

Boyle also has a particular expertise in regulated industries and has joined us for previous events to lay out some advice for startups navigating governmental rules.

On Tuesday, we’ll discuss trends and industries they find most exciting as we head into 2021, get their best advice for early-stage startups seeking funding and hear how the VC landscape has changed during the pandemic.

You should most definitely bring your own questions to the table. Extra Crunch members can submit questions ahead of time via the form below or during the live discussion.

Full details can be found below.

See you there!

Event Details

Last year, AWS launched the APN Global Startup Program, which is sort of AWS’s answer to an incubator for startups deeply involved with AWS technology. This year, the company wants to expand that offering, and today it announced some updates to the program at the Partner keynote today at AWS re:Invent.

While startups technically have to pay a $2500 fee if they are accepted to the program, AWS typically refunds that fee, says Doug Yeum, head of the Global Partner Organization at AWS — and they get a lot of benefits for being part of the program.

“While the APN has a $2,500 annual program fee, startups that are accepted into the invite-only APN Global Startup Program get that fee back, as well as free access to substantial additional resources both in terms of funding as well as exclusive program partner managers and co-sell specialists resources,” Yeum told TechCrunch.

And those benefits are pretty substantial including access to a new “white glove program” that lets them work with a program manager with direct knowledge of AWS and who has experience working with startups. In addition, participants get access to an ISV program to work more directly with these vendors to increase sales and access to data exchange services to move third party data into the AWS cloud.

What’s more, they can apply to the new AI/ML Acceleration program. As AWS describes it, “This includes up to $5,000 AWS credits to fund experiments on AWS services, enabling startups to explore AWS AI/ML tools that offer the best fit for them at low risk.”

Finally, they get partially free access to the AWS Marketplace, offsetting the normal marketplace listing fees for the first five offerings. Some participants will also get access to AWS sales to help use the power of the large company to drive a startup’s sales.

While you can apply to the program, the company also recruits individual startups that catch its attention. “We also proactively invite mid-to-late stage startups built on AWS that, based on market signals, are showing traction and offer interesting use cases for our mutual enterprise customers,” Yeum explained.

Among the companies currently involved in the program are HashiCorp, Logz.io and Snapdocs. Interested startups can apply on the APN Global Startup website.

Last year, AWS launched the APN Global Startup Program, which is sort of AWS’s answer to an incubator for startups deeply involved with AWS technology. This year, the company wants to expand that offering, and today it announced some updates to the program at the Partner keynote today at AWS re:Invent.

While startups technically have to pay a $2500 fee if they are accepted to the program, AWS typically refunds that fee, says Doug Yeum, head of the Global Partner Organization at AWS — and they get a lot of benefits for being part of the program.

“While the APN has a $2,500 annual program fee, startups that are accepted into the invite-only APN Global Startup Program get that fee back, as well as free access to substantial additional resources both in terms of funding as well as exclusive program partner managers and co-sell specialists resources,” Yeum told TechCrunch.

And those benefits are pretty substantial including access to a new “white glove program” that lets them work with a program manager with direct knowledge of AWS and who has experience working with startups. In addition, participants get access to an ISV program to work more directly with these vendors to increase sales and access to data exchange services to move third party data into the AWS cloud.

What’s more, they can apply to the new AI/ML Acceleration program. As AWS describes it, “This includes up to $5,000 AWS credits to fund experiments on AWS services, enabling startups to explore AWS AI/ML tools that offer the best fit for them at low risk.”

Finally, they get partially free access to the AWS Marketplace, offsetting the normal marketplace listing fees for the first five offerings. Some participants will also get access to AWS sales to help use the power of the large company to drive a startup’s sales.

While you can apply to the program, the company also recruits individual startups that catch its attention. “We also proactively invite mid-to-late stage startups built on AWS that, based on market signals, are showing traction and offer interesting use cases for our mutual enterprise customers,” Yeum explained.

Among the companies currently involved in the program are HashiCorp, Logz.io and Snapdocs. Interested startups can apply on the APN Global Startup website.

At our fast-approaching first TC Sessions: Space event, which is happening December 16-17, we’re going to be highlighting some of the most exciting startups and founders tackling big problems with innovative and groundbreaking solutions.

Some of those companies are focused on building tomorrow’s spacecraft, and others are working on in-space technologies that could become the next big anchor upon which countless other businesses are built.

Two of the companies joining us at TC Sessions: Space are Skyroot and Bluefield. Skyroot is India’s first private space launch startup, founded in 2018 with the goal of developing a low-cost and reliable launch vehicle to help democratize access of space.

Founder, CEO and CTO Pawan Kumar Chandana will join us to talk about building his new business, his prior experience developing rockets for the Indian Space Research Organization (ISRO) and how Skyroot’s Vikram-series launch vehicles plan to achieve the company’s ambitious goals.

Bluefield Technologies is focused on an entirely different, but potentially just as impactful opportunity: Observation, monitoring and analysis of methane emissions data on Earth. Their satellite-based methane observation technology offer a new high bar of precision and detail.

Bluefield founder and CEO Yotam Ariel will join us to talk about what becomes possible across a range of industries once you offer them the ability to track up to 90% of the Earth’s methane emissions with pinpoint accuracy, at costs that are up to 100 percent cheaper than existing solutions on up to a daily basis.

We’ll have conversations with Chandana, Ariel and others as part of our ‘Founders in Focus’ series, just one small part of the all-star agenda at TC Sessions: Space. Tickets are still available at the Late Registration price with discounts for students, government/military employees and groups, so grab yours below to attend this fully virtual event.

At our fast-approaching first TC Sessions: Space event, which is happening December 16-17, we’re going to be highlighting some of the most exciting startups and founders tackling big problems with innovative and groundbreaking solutions.

Some of those companies are focused on building tomorrow’s spacecraft, and others are working on in-space technologies that could become the next big anchor upon which countless other businesses are built.

Two of the companies joining us at TC Sessions: Space are Skyroot and Bluefield. Skyroot is India’s first private space launch startup, founded in 2018 with the goal of developing a low-cost and reliable launch vehicle to help democratize access of space.

Founder, CEO and CTO Pawan Kumar Chandana will join us to talk about building his new business, his prior experience developing rockets for the Indian Space Research Organization (ISRO) and how Skyroot’s Vikram-series launch vehicles plan to achieve the company’s ambitious goals.

Bluefield Technologies is focused on an entirely different, but potentially just as impactful opportunity: Observation, monitoring and analysis of methane emissions data on Earth. Their satellite-based methane observation technology offer a new high bar of precision and detail.

Bluefield founder and CEO Yotam Ariel will join us to talk about what becomes possible across a range of industries once you offer them the ability to track up to 90% of the Earth’s methane emissions with pinpoint accuracy, at costs that are up to 100 percent cheaper than existing solutions on up to a daily basis.

We’ll have conversations with Chandana, Ariel and others as part of our ‘Founders in Focus’ series, just one small part of the all-star agenda at TC Sessions: Space. Tickets are still available at the Late Registration price with discounts for students, government/military employees and groups, so grab yours below to attend this fully virtual event.

VCs expect the companies they invest in to use data to improve their decision-making. So why aren’t they doing that when evaluating startup teams?

Sure, venture capital is a people business, and the power of gut feeling is real. But using an objective, data-backed process to evaluate teams — the same way we do when evaluating financial KPIs, product, timing and market opportunities — will help us make better investment decisions, avoid costly mistakes and discover opportunities we might have otherwise overlooked.

An objective assessment process will also help investors break free from patterns and back someone other than a white male for a change. Is looking at how we have always done things the best way to build for the future?

Sixty percent of startups fail because of problems with the team. Instinct matters, but a team is too big a risk to leave to intuition. I will use myself as an example. I have founded two companies. I know what it takes to build a company and to achieve a successful exit. I like to think I can sense when someone has that special something and when a team has chemistry. But I am human. I am limited by bias and thought patterns; data is not.

You can (and should) take a scientific approach to evaluating a startup team. A “strong” team isn’t a vague concept — extensive research confirms what it takes to execute a vision. Despite what people expect, soft skills can be measured. VCVolt is a computerized selection model that analyzes the performance of companies and founding teams developed by Eva de Mol, Ph.D., my partner at CapitalT.

We use it to inform every investment decision we make and to demystify a common hurdle to entrepreneurial success. (The technology also evaluates the company, market opportunity, timing and other factors, but since most investors aren’t taking a structured, data-backed approach to analyzing teams, let’s focus on that.)

VCVolt allows us to reduce team risk early on in the selection and due diligence process, thereby reducing confirmation bias and fail rates, discovering more winning teams and driving higher returns.

I will keep this story brief for privacy reasons, but you will get the point. While testing the model, we advised another VC firm not to move forward with an investment based on the model’s findings. The firm moved forward anyway because they were in love with the deal, and everything the model predicted transpired. It was a big loss for the investors, and a reminder that hunch and gut feeling can be wrong — or at least blind you to some serious risk factors.

The platform uses a validated model that is based on more than five years of scientific research, data from more than 1,000 companies and input from world-class experts and scientists. Its predictive validity is noted in top-tier scientific journals and other publications, including Harvard Business Review. By asking the right questions — science-based questions validated by more than 80,000 datapoints — the platform analyzes the likelihood that a team will succeed. It considers:

Atlassian is launching an update to its ubiquitous Jira issue and project tracking service today — and specifically the Jira Software Cloud version — that brings a number of new features for visualizing and measuring how code moves through the development pipeline. With this, project managers and developers will be able to get deeper insights into the code teams are working on, for example, and where that code is in the deployment pipeline. Users of Jira Software Premium will also be able to track deployment frequency and cycle times right inside of the service now, too.

“In the quest to reach that near-mythical land of ‘insights,’ many teams mistake consolidation for control. But the challenge is not the number of tools; it’s the way they’re integrated,” the Jira team writes in today’s announcement. “No single vendor will ever deliver all the products an agile software team needs, so the burden still lies on the team to manually connect the dots.”

Image Credits: Atlassian

This update, Atlassian argues, helps those teams do just that. For most teams, Jira is already the central repository where each piece of work is documented in some form or another, after all. Some of that work happens in Atlassian tools, but most of it happens in the context of third-party services. The idea here is to pull all of this DevOps work together and provide more visibility and insights into the state of a company’s development pipeline.

Specifically, there are four different updates here. The first is ‘code in Jira,’ which may sound like you can now code inside of Jira, but in reality, it’s about seeing which repos in Bitbucket, GitHub, GitLab or Git Integration for Jira are currently actively worked on. With the new ‘deployments’ feature, users can now get a real-time view of all of their deployment information across CI/CD services like Bitbucket Pipelines, Jenkins, Azure DevOps, Circle CI, Octopus Deploy and JFrog.

Image Credits: Atlassian

“Whether you’re a product manager looking to see which features have deployed to which environment or a team lead looking to understand the average time it takes for your team to go from idea to production within a certain project, you’ll find your answer in the Deployments in Jira tab,” the company explains.

The last two features, only available in the pricier and more enterprisey Jira Software Premium, will soon provide more in-depth metrics about deployment frequency and cycle times. The idea here is simply to provide more metrics to help teams better understand trends and identify outliers in their processes.

If you want to make a machine learning system, you need data for it, but that data isn’t always easy to come by. MLCommons aims to unite disparate companies and organizations in the creation of large public databases for AI training, so that researchers around the world can work together at higher levels, and in doing so advance the nascent field as a whole. Its first effort, the People’s Speech dataset, is many times the size of others like it, and aims to be more diverse as well.

MLCommons is a new non-profit related to MLPerf, which has collected input from dozens of companies and academic institutions to create industry-standard benchmarks for machine learning performance. The endeavor has met with success, but in the process the team encountered a paucity of open datasets that everyone could use.

If you want to do an apples-to-apples comparison of a Google model to an Amazon model, or for that matter a UC Berkeley model, they really all ought to be using the same testing data. With computer vision one of the most widespread datasets is ImageNet, which is used and cited by all the most influential papers and experts. But there’s no such dataset for, say, speech to text accuracy.

“Benchmarks get people talking about progress in a sensible, measurable way. And it turns out that if the goal is the move the industry forward, we need datasets we can use — but lots of them are difficult to use for licensing reasons, or aren’t state of the art,” said MLCommons co-founder and executive director David Kanter.

Certainly the big companies have enormous voice datasets of their own, but they’re proprietary and perhaps legally restricted from being used by others. And there are public datasets, but with only a few thousand hours their utility is limited — to be competitive today one needs much more than that.

“Building large datasets is great because we can create benchmarks, but it also moves the needle forward for everyone. We can’t rival what’s available internally but we can go a long way towards bridging that gap,” Kanter said. MLCommons is the organization they formed to create and wrangle the required data and connections.

The People’s Speech Dataset was assembled from a variety of sources, with about 65,000 of its hours coming from audiobooks in English, with the text aligned with the audio. Then there are 15,000 hours or so sourced from around the web, with different acoustics, speakers, and styles of speech (for example conversational instead of narrative). 1,500 hours of English audio were sourced from Wikipedia, and then 5,000 hours of synthetic speech of text generated by GPT-2 were mixed in (“A little bit of the snake eating its own tail,” joked Kanter). 59 languages in total are represented in some way, though as you can tell it is mostly English.

Although diversity is the goal — you can’t build a virtual assistant in Portuguese from English data — it’s also important to establish a baseline for what’s needed for present purposes. Is 10,000 hours sufficient to build a decent speech-to-text model? Or does having 20,000 available make development that much easier, faster, or effective? What if you want to be excellent at American English but also decent with Indian and English accents? How much of those do you need?

The general consensus with datasets is simply “the larger the better,” and the likes of Google and Apple are working with far more than a few thousand hours. Thus the 86,000 hours in this first iteration of the dataset. And it is definitely the first of many, with later versions due to branch out into more languages and accents.

“Once we verify we can deliver value, we’ll just release and be honest about the state it’s in,” explained Peter Mattson, another co-founder of MLCommons and currently head of Google’s Machine Learning Metrics Group. “We also need to learn how to quantify the idea of diversity. The industry wants this; we need more dataset construction expertise — there’s tremendous ROI for everybody in supporting such an organization.”

The organization is also hoping to spur sharing and innovation in the field with MLCube, a new standard for passing models back and forth that takes some of the guesswork and labor out of that process. Although machine learning is one of the tech sector’s most active areas of research and development, taking your AI model and giving to someone else to test, run, or modify isn’t as simple as it ought to be.

Their idea with MLCube is a wrapper for models that describes and standardizes a few things, like dependencies, input and output format, hosting and so on. AI may be fundamentally complex, but it and the tools to create and test it are still in their infancy.

The dataset should be available now, or soon, from MLCommons’ website, under the CC-BY license, allowing for commercial use; a few reference models trained on the set will also be released.