Category: UNCATEGORIZED

Have you ever wanted to hear a concerto for piano and harp, in the style of Mozart by way of Katy Perry? Well, why not? Because now you can, with OpenAI’s latest (and blessedly not potentially catastrophic) creation, MuseNet. This machine learning model produces never-before-heard music basic on its knowledge of artists and a few bars to fake it with.

This is far from unprecedented — computer-generated music has been around for decades — but OpenAI’s approach appears to be flexible and scalable, producing music informed by a variety of genres and artists, and cross-pollinating them as well in a form of auditory style transfer. It shares a lot of DNA with GPT2, the language model “too dangerous to release,” but the threat of unleashing unlimited music on the world seems small compared with undetectable computer-generated text.

MuseNet was trained on works from dozens of artists, from well-known historical figures like Chopin and Bach to (comparatively) modern artists like Adele and the Beatles, plus collections of African, Arabic, and Indian music. Its complex machine learning system paid a great deal of “attention,” which is a technical term in AI work for, essentially, the amount of context the model uses to inform the next step in its creation.

Take, for instance, a piece by Mozart. If the model only attended to a couple seconds at a time, it would never be able to learn the larger musical structures of a symphony as it grew and receded, switched tones and instruments. But the model was given enough virtual brainspace to hold onto about four full minutes of sound, more than enough to grasp something like a slow start to a big finish, or a basic verse-chorus-verse structure.

You’re telling me Haydn didn’t directly influence Shania? Get real.

Theoretically, that is. The model doesn’t actually understand music theory, just that this note followed this note, which followed this note, which tends to come after this type of chord, and so on. Its creations are elementary in their structure, but it’s pretty clear listening to them that it is indeed successfully aping the songs it ingested.

What makes it impressive is that a single model does this reliably across so many types of music. AIs have been created, like the Google Doodle for Bach’s birthday a couple weeks back, that focus on a specific artist or genre. And as a comparison I’ve been listening to Generative.fm, which creates just the type of sparse ambient music I like to listen to while I work (If you like it too, check out one of my favorite labels, Serein). But both those models have their limits very strictly defined. Not so with MuseNet.

In addition to being able to belt out infinite bluegrass or baroque piano pieces, MuseNet can apply a style transfer process to combine the characteristics of both. Different parts of a work can have different attributes — in a painting you might have composition, subject, color choice, and brush style to start. Imagine a Pre-Raphaelite subject and composition but with Impressionist execution. Sounds fun, right? AI models are great at doing this because they sort of compartmentalize these different aspects. It’s the same type of thing in music: The note choice, cadence, and other patterns of a pop song can be drawn out and used separately from its instrumentation — why not do Beach Boys harmonies on a harp?

It’s a little hard, however, to get a sense of the likes of Adele without her distinctive voice, and the rather basic synths the team has chosen cheapen the effect overall. And after listening to the “live concert” the team gave on Twitch for a bit, I wasn’t convinced that MuseNet is the next hit machine. On the other hand, it pretty regularly hit a good stride, especially in jazz and classical improvisations, where a bit of an off note can be played off and the rhythms don’t feel so contrived.

What’s it for? Your idea is as good as anyone’s, really. This field is quite new. MuseNet’s project lead, Christine Payne, is pleased with the model and has already found someone to use it:

As a classically trained pianist, I’m particularly excited to see that MuseNet is able to understand the complex harmonic structures of Beethoven and Chopin. I’m working now with a composer who plans to integrate MuseNet into his own compositions, and I’m excited to see where the future of Human/AI co-composing will take us.

MuseNet will be available for you to play with through mid-May, at which point it will be taken offline and adjusted based on feedback from users, and later it will be at least partially open sourced. I imagine popular combinations and ones that people listened to all the way through will get a bit more weight in the tweaks. Here’s hoping they add a bit more expression to the MIDI execution as well — it does often feel like these pieces are being played by a robot. But it’s testament to the quality of OpenAI’s work that they frequently sound perfectly good as well.

Have you ever wanted to hear a concerto for piano and harp, in the style of Mozart by way of Katy Perry? Well, why not? Because now you can, with OpenAI’s latest (and blessedly not potentially catastrophic) creation, MuseNet. This machine learning model produces never-before-heard music basic on its knowledge of artists and a few bars to fake it with.

This is far from unprecedented — computer-generated music has been around for decades — but OpenAI’s approach appears to be flexible and scalable, producing music informed by a variety of genres and artists, and cross-pollinating them as well in a form of auditory style transfer. It shares a lot of DNA with GPT2, the language model “too dangerous to release,” but the threat of unleashing unlimited music on the world seems small compared with undetectable computer-generated text.

MuseNet was trained on works from dozens of artists, from well-known historical figures like Chopin and Bach to (comparatively) modern artists like Adele and the Beatles, plus collections of African, Arabic, and Indian music. Its complex machine learning system paid a great deal of “attention,” which is a technical term in AI work for, essentially, the amount of context the model uses to inform the next step in its creation.

Take, for instance, a piece by Mozart. If the model only attended to a couple seconds at a time, it would never be able to learn the larger musical structures of a symphony as it grew and receded, switched tones and instruments. But the model was given enough virtual brainspace to hold onto about four full minutes of sound, more than enough to grasp something like a slow start to a big finish, or a basic verse-chorus-verse structure.

You’re telling me Haydn didn’t directly influence Shania? Get real.

Theoretically, that is. The model doesn’t actually understand music theory, just that this note followed this note, which followed this note, which tends to come after this type of chord, and so on. Its creations are elementary in their structure, but it’s pretty clear listening to them that it is indeed successfully aping the songs it ingested.

What makes it impressive is that a single model does this reliably across so many types of music. AIs have been created, like the Google Doodle for Bach’s birthday a couple weeks back, that focus on a specific artist or genre. And as a comparison I’ve been listening to Generative.fm, which creates just the type of sparse ambient music I like to listen to while I work (If you like it too, check out one of my favorite labels, Serein). But both those models have their limits very strictly defined. Not so with MuseNet.

In addition to being able to belt out infinite bluegrass or baroque piano pieces, MuseNet can apply a style transfer process to combine the characteristics of both. Different parts of a work can have different attributes — in a painting you might have composition, subject, color choice, and brush style to start. Imagine a Pre-Raphaelite subject and composition but with Impressionist execution. Sounds fun, right? AI models are great at doing this because they sort of compartmentalize these different aspects. It’s the same type of thing in music: The note choice, cadence, and other patterns of a pop song can be drawn out and used separately from its instrumentation — why not do Beach Boys harmonies on a harp?

It’s a little hard, however, to get a sense of the likes of Adele without her distinctive voice, and the rather basic synths the team has chosen cheapen the effect overall. And after listening to the “live concert” the team gave on Twitch for a bit, I wasn’t convinced that MuseNet is the next hit machine. On the other hand, it pretty regularly hit a good stride, especially in jazz and classical improvisations, where a bit of an off note can be played off and the rhythms don’t feel so contrived.

What’s it for? Your idea is as good as anyone’s, really. This field is quite new. MuseNet’s project lead, Christine Payne, is pleased with the model and has already found someone to use it:

As a classically trained pianist, I’m particularly excited to see that MuseNet is able to understand the complex harmonic structures of Beethoven and Chopin. I’m working now with a composer who plans to integrate MuseNet into his own compositions, and I’m excited to see where the future of Human/AI co-composing will take us.

MuseNet will be available for you to play with through mid-May, at which point it will be taken offline and adjusted based on feedback from users, and later it will be at least partially open sourced. I imagine popular combinations and ones that people listened to all the way through will get a bit more weight in the tweaks. Here’s hoping they add a bit more expression to the MIDI execution as well — it does often feel like these pieces are being played by a robot. But it’s testament to the quality of OpenAI’s work that they frequently sound perfectly good as well.

The Markup appears to be facing a staff revolt — and its financial backers may be reconsidering their support — following the firing of Editor in Chief Julia Angwin.

When the site was announced last fall, it was backed by $20 million from Craigslist founder Craig Newmark, with additional funding from the John S. and James L. Knight Foundation, the Ford Foundation and the John D. and Catherine T. MacArthur Foundation. The goal was to do data-driven journalism about the impact of technology on society.

Angwin and her co-founder Jeff Larson seemed particularly well-suited for the job — both of them are award-winning journalists who worked together at ProPublica, where they did impactful reporting around topics like Facebook’s ad practices.

However, Angwin was fired on Monday, a move she blamed in interviews on executive director Sue Gardner’s plan to turn the site into “a cause, not a publication,” with headlines like “Facebook is a dumpster fire.”

This, Angwin said, was at odds with her own dedication to “evidence-based, data-driven journalism.”

Larson, who’s now become editor-in-chief, offered a different account on Medium, where he said work had fallen “far, far behind” by the end of 2018: “Hiring was slow. Recruitment was slow. Even as of this month, we didn’t have stories banked. We didn’t have editorial processes in place to accept and develop pieces.”

He said that he and Angwin were both asked to take management classes, but she refused. (Angwin acknowledged that she may have had things to learn about being editor-in-chief, but she noted that she’s led investigative teams in the past, and she said, “There was never any attempt to guide me into that learning.”)

Larson also alluded to other issues that led to “a breakdown in trust between the three of us as co-founders.” He said there were attempts to find other roles for Angwin, but she “refused to discuss any role other than Editor in Chief, and would not consider any other configuration. So unfortunately we made the decision to remove her from that role.”

The editorial team has sided with  Angwin, with all of them posting a statement supporting her and praising her “effectiveness as a manager and an editor.” Five of the seven editorial team members also resigned in protest.

As a result of all the controversy, Newmark and the other funders of The Markup have issued a statement of their own, saying that while they’re still “committed to the mission of The Markup,” they’ve also decided “it is necessary to reassess our support and we are taking steps to do so.”

Apple is rolling out a fix for an App Store bug that was preventing users from downloading new iOS apps or app updates. The issue, which impacted an unknown number of users, involved a Terms & Conditions dialog box that would continue to pop up even when users tapped the “Agree” button.

The issue had frustrated users who took to Twitter in an attempt to get help from Apple Support.

9to5Mac and AppleInsider previously reported on the problem, citing the social media complaints. The Apple Support account had not responded publicly to those who reached out, beyond asking customers to get in touch on DM with more details or pointing those with more vague complaints to a support doc about connection issues.

The bug was affecting a small percentage of Apple’s iOS user base worldwide, we understand from people familiar with the problem at Apple. However, even a “small number” can be large, when you’re dealing with something like the iPhone install base.

In addition, the bug wasn’t limited to an unreleased developer or beta build of iOS, 9to5Mac had noted — but was also showing up on the public release (iOS 12.2).

There was no workaround that would allow users to skip the Terms & Condition pop-up in order to download apps and updates. All users could do was to tap “Cancel” to get out of the loop, so they could continue to use their phone.

We understand that Apple has now deployed a fix for the bug that should finish rolling out in a matter of hours.

The bug fix won’t require impacted users to take any steps on their own — like downloading an update or patch, for example. It will instead be resolved on the App Store’s backend.

Amazon is always looking for ways to increase the options it offers developers in AWS, and to that end, today it announced a bunch of new AMD EPYC-powered T3a instances. These were originally announced at the end of last year at re:Invent, AWS’s annual customer conference.

Today’s announcement is about making these chips generally available. They have been designed for a specific type of burstable workload, where you might not always need a sustained amount of compute power.

“These instances deliver burstable, cost-effective performance and are a great fit for workloads that do not need high sustained compute power but experience temporary spikes in usage. You get a generous and assured baseline amount of processing power and the ability to transparently scale up to full core performance when you need more processing power, for as long as necessary,” AWS’s Jeff Barr wrote in a blog post.

These instances are build on the AWS Nitro System, Amazon’s custom networking interface hardware that the company has been working on for the last several years. The primary components of this system include the Nitro Card I/O Acceleration, Nitro Security Chip and the Nitro Hypervisor.

Today’s release comes on top of the announcement last year that the company would be releasing EC2 instances powered by Arm-based AWS Graviton Processors, another option for developers, who are looking for a solution for scale-out workloads.

It also comes on the heels of last month’s announcement that it was releasing EC2 M5 and R5 instances, which use lower-cost AMD chips. These are also built on top of the Nitro System.

The EPCY processors are available starting today in seven sizes in your choice of spot instances, reserved instances or on-demand, as needed. They are available in US East in northern Virginia, US West in Oregon, Europe in ireland, US East in Ohio and Asia-Pacific in Singapore.

If you’re a student at UC Berkeley, the diminutive rolling robots from Kiwi are probably a familiar sight by now, trundling along with a burrito inside to deliver to a dorm or apartment building. Now students at a dozen more campuses will be able to join this great, lazy future of robotic delivery as Kiwi expands to them with a clever student-run model.

Speaking at TechCrunch’s Robotics/AI Session at the Berkeley campus, Kiwi’s Felipe Chavez and Sasha Iatsenia discussed the success of their burgeoning business and the way they planned to take it national.

In case you’re not aware of the Kiwi model, it’s basically this: When you place an order online with a participating restaurant, you have the option of delivery via Kiwi. If you so choose, one of the company’s fleet of knee-high robots with insulated, locking storage compartments will swing by the place, your order is put within, and it brings it to your front door (or as close as it can reasonably get). You can even watch the last bit live from the robot’s perspective as it rolls up to your place.

The robots are what Kiwi calls “semi-autonomous.” This means that although they can navigate most sidewalks and avoid pedestrians, each has a human monitoring it and setting waypoints for it to follow, on average every five seconds. Iatsenia told me that they’d tried going full autonomous and that it worked… most of the time. But most of the time isn’t good enough for a commercial service, so they’ve got humans in the loop. They’re working on improving autonomy but for now this is how it is.

That the robots are being controlled in some fashion by a team of people in Colombia (where the co-founders hail from) does take a considerable amount of the futurism out of this endeavor, but on reflection it’s kind of a natural evolution of the existing delivery infrastructure. After all, someone has to drive the car that brings you your food as well. And in reality most AI is operated or informed directly or indirectly by actual people.

That those drivers are in South America operating multiple vehicles at a time is a technological advance over your average delivery vehicle — though it must be said that there is an unsavory air of offshoring labor to save money on wages. That said, few people shed tears over the wages earned by the Chinese assemblers who put together our smartphones and laptops, or the garbage pickers who separate your poorly sorted recycling. The global labor economy is a complicated one, and the company is making jobs in the place it was at least partly born.

Whatever the method, Kiwi has traction: it’s done more than 50,000 deliveries and the model seems to have proven itself. Customers are happy, they get stuff delivered more than ever once they get the app, and there are fewer and fewer incidents where a robot is kicked over or, you know, catches on fire. Notably, the founders said on stage, the community has really adopted the little vehicles, and should one overturn or be otherwise interfered with, it’s often set on its way soon after by a passerby.

Iatsenia and Chavez think the model is ready to push out to other campuses, where a similar effort will have to take place — but rather than do it themselves by raising millions and hiring staff all over the country, they’re trusting the robotics-loving student groups at other universities to help out.

For a small and low-cash startup like Kiwi, it would be risky to overextend by taking on a major round and using that to scale up. They started as robotics enthusiasts looking to bring something like this to their campus, so why can’t they help others do the same?

So the team looked at dozens of universities, narrowing them down by factors important to robotic delivery: layout, density, commercial corridors, demographics, and so on. Ultimately they arrived at the following list:

• Northern Illinois University
• University of Oklahoma
• Purdue University
• Texas A&M
• Parsons
• Cornell
• East Tennessee State University
• Nebraska University-Lincoln
• Stanford
• Harvard
• NYU
• Rutgers

What they’re doing is reaching out to robotics clubs and student groups at those colleges to see who wants to take partial ownership of Kiwi administration out there. Maintenance and deployment would still be handled by Berkeley students, but the student clubs would go through a certification process and then do the local work, like a capsized bot and on-site issues with customers and restaurants.

“We are exploring several options to work with students down the road including rev share,” Iatsenia told me. “It depends on the campus.”

So far they’ve sent out 40 robots to the 12 campuses listed and will be rolling out operations as the programs move forward on their own time. If you’re not one of the unis listed, don’t worry — if this goes the way Kiwi plans, it sounds like you can expect further expansion soon.

Social media-fuelled anti-vaxxer propaganda is the latest online harm the U.K. government is targeting.

Speaking on BBC Radio 4’s Today program this morning health secretary Matt Hancock said he will meet with representatives from social media platforms on Monday to pressure them into doing more to prevent false information about the safety of vaccinations from being amplified by their platforms.

“I’m seeing them on Monday to require that they do more to take down wrong — well lies essentially — that are promoted on social media about the impact of vaccination,” he said, when asked about a warning by a U.K. public health body about the risk of a public health emergency being caused by an increase in the number of British children who have not received the measles vaccination.

“Vaccination is safe; it’s very, very important for the public health, for everybody’s health and we’re going to tackle it.”

The head of NHS England also warned last month about anti-vaccination messages gaining traction on social media.

“We need to tackle this risk in people not vaccinating,” Hancock added. “One of the things I’m particularly worried about is the spread of anti-vaccination messages online. I’ve called in the social media companies like we had to for self-harming imagery a couple of months ago.”

Hancock, who between 2016 and 2018 served as the U.K.’s digital minister, prior to taking over the health brief, held a similar meeting with the boss of Instagram earlier this year.

That followed a public outcry over suicide content spreading on Instagram after a British schoolgirl was reported to have been encouraged to killed herself by graphic content on the Facebook -owned platform.

Instagram subsequently announced a policy change saying it would remove graphic images of self harm and remove non-graphic self-harm images so they don’t show up in searches, relevant hashtags or the explore tab.

But it remains to be seen whether platforms will be as immediately responsive to amped up political pressure to scrub anti-vaccination content entirely given the level of support anti-vaxxer messages can attract among social media users.

Earlier this year Facebook said it would downrank such content in the News Feed and hide it on Instagram in an effort to minimize the spread of vaccination misinformation.

It also said it would point users toward “authoritative” vaccine-related information — i.e. information that’s been corroborated by the health and scientific establishment.

But deleting such content entirely was not part of Facebook’s announced strategy.

We’ve reached out to Facebook for any response to Hancock’s comments.

In the longer term social media platforms operating in the U.K. could face laws that require them to remove content deemed to pose a risk to public health if ordered to by a dedicated regulator, as a result of a wide-ranging government plan to tackle a range of online harms.

Earlier this month the U.K. government set out a broad policy plan for regulating online harms.

The Online Harms Whitepaper proposes to put a mandatory duty of care on platforms to take reasonable steps to protect users from a range of harms — including those linked to the spread of disinformation.

It also proposes a dedicated, overarching regulator to oversee internet companies to ensure they meet their responsibilities.

The government is currently running a public consultation on the proposals, which ends July 1, after which it says it will set out any next actions as it works on developing draft legislation.

Huawei’s controversial status in North America and Europe stems from a lot of different factors. At the heart of most of them, however, is the hardware maker’s alleged ties to the Chinese government. The notion of government control has been enough to cause something approaching an outright ban on its products in the States, over worries that handsets and networking equipment could be used to spy on the U.S. government and its citizens.

A recently published report from professors at Fulbright University Vietnam and George Washington University Law School resurfaced those issues. The simply titled “Who Owns Huawei?” attempts to get to the bottom of who is controlling the rapidly ascending smartphone maker. The results published struggle to draw a clear conclusion on the matter, though the authors note in the summary, “Regardless of who, in a practical sense, owns and controls Huawei, it is clear that the employees do not.”

Huawei held a press call this week in an attempt to clarify some of that confusion, and the results were, well, also pretty confusing. The company provided TechCrunch with a transcript of the remarks by Chief Secretary of its Board of Directors, Jiang Xisheng. Jiang explained that, contrary to reports, Huawei is “wholly owned by its employees.”

The executive is referring to Huawei’s labor union. Under the plan, employees control 99 percent of the company’s “virtual restricted shares.” The executive explains the structure thusly:

In China, a limited liability company can have up to 50 registered shareholders. A non-listed stock corporation can have up to 200 registered shareholders. At Huawei, we have way more than 50 or 200 shareholding employees, so they cannot be registered as Huawei’s shareholders. This is true for Huawei as a limited liability company. Even if we make our company a stock corporation, it would still be impossible to register all our shareholding employees as shareholders. Because of this, the Union acts as a platform through which our employees can hold shares.

Certainly sounds nice, but as The Wall Street Journal notes, founder Ren Zhengfei only has one percent, but makes the key decisions, including who sits on the board and other major moves. “The Trade Union Committee also does not influence the operations of Huawei Holding or Huawei Technologies,” Jiang explains. “The Trade Union Committee is not involved in any of the company’s business operations.”

Jiang says the trade union involves itself with improving the physical and mental well-being of its staff, from helping to pay for medical expenses to organizing a variety of clubs, including basketball and badminton.

The paper, however, dismisses the notion that trade union ownership and government control are mutually exclusive. “Given the public nature of trade unions in China,” its authors write, “if the ownership stake of the trade union committee is genuine, and if the trade union and its committee function as trade unions generally function in China, then Huawei may be deemed effectively state-owned.”

Jiang, for his part, outright rejects the notion of government stake in the company. “Most of what the US government says is not true,” he says. “Regarding this point, we have responded many times. Though it is not under my charge, one thing is for sure – there is no government capital in Huawei. Huawei issued some bonds, many in the capital markets in Hong Kong and in countries outside of China. So far, to my knowledge, we have not issued bonds on the Chinese mainland.”

The Daily Crunch is TechCrunch’s roundup of our biggest and most important stories. If you’d like to get this delivered to your inbox every day at around 9am Pacific, you can subscribe here.

1. Facebook reserves $3B for FTC fine, but keeps growing with 2.38B users in Q1

A massive penalty hangs over Facebook’s head, but it otherwise had a very strong Q1 earnings report. The company reached 2.38 billion monthly users, up 2.5 percent from the previous quarter, and it pulled in $15.08 billion in revenue.

Facebook recorded earnings per share were significantly lower than expected, but that’s because it set aside $3 billion to cover a potential FTC fine that it’s still resolving.

2. Scientists pull speech directly from the brain

In a feat that could eventually unlock the possibility of speech for people with severe medical conditions, scientists have successfully recreated the speech of healthy subjects by tapping directly into their brains.

3. Verizon announces 20 5G markets for 2019, as Samsung Galaxy S10 5G preorders open

Verizon (which owns TechCrunch) just revealed a one-two punch: opening up preorders for the Galaxy S10 5G and announcing a list of 20 cities that will be getting the technology before year’s end.

4. Microsoft beats expectations with $30.6B in revenue as Azure’s growth continues

Microsoft Azure had a pretty good quarter, with revenue growing 73 percent. That’s a bit lower than last quarter’s results, but only by a fraction.

5. Slack to extend collaboration to folks who don’t want to give up email

The company announced a new email and calendar bridge that enables team members who might not have made the leap to Slack to still be kept in the loop.

6. NASA and FEMA are contingency planning for a potential asteroid Armageddon

Alongside international partners, NASA’s Planetary Defense Coordination Office will participate in a “tabletop exercise” that will simulate a scenario for how to respond to an asteroid on an impact trajectory with Earth.

7. How to source hard-to-fill programming positions

Zack Burt’s recruiting strategy is surprisingly simple, and boils down to optimizing various segments of the sourcing funnel: awareness, page views and application submits. (Extra Crunch membership required.)