Qualcomm unveils the Snapdragon 8c, with a Kryo 490 CPU, Adreno 675 GPU, and X24 LTE modem, and a lower-end 7c chip; both designed for budget Windows laptops — The Snapdragon 8c and 7c join the high-end 8cx — Qualcomm has had big ambitions for ARM-powered Windows laptops for years.
Cyberhaven, which provides data behavior analytics to protect company trade secrets such as IP for customers including Motorola and DARPA, raises $13M Series A — For obvious reasons, trade secrets are highly valuable — they're the innovations that confer competitive advantage.
Uber’s just-released U.S. Safety Report sets forth in some detail the number of fatal accidents, and the good news is that the overall rate per mile is about half the national average. But the report makes some puzzling choices as far as what is included and excluded.
To create the report, Uber took its internal reports of crashes, generated by drivers, users, or insurance companies, and compared it to the national Fatality Analysis Reporting System, or FARS, a database that tracks all automotive deaths. In this way Uber was able to confirm 97 fatal crashes with 107 total deaths in 2017 and 2018 combined.
As the company is careful to point out before this, more than 36,000 people died in car crashes in the U.S. in 2018 alone, so the total doesn’t really mean much on its own. So they (as others do in this field) put those accidents in context of miles traveled. After all, 1 crash in 100,000 miles doesn’t sound bad because it’s only one, but 10 crashes in a billion miles, which is closer to what Uber saw, is actually much better despite the first number being higher. To some this is blindingly obvious but perhaps not to others.
The actual numbers are that in 2017, there were 49 “Uber-related” fatalities over 8.2 billion miles, or approximately 0.59 per 100 million miles traveled; in 2018, there were 58 over 1.3 billion, or about 0.57 per 100 million miles. The national average is more than 1.1 per 100 million, so Uber sees about half as many fatalities per mile overall.
These crashes generally occurred at lower speeds than the national average, and were more likely by far to occur at night, in lighted areas of cities. That makes sense, since rideshare services are heavily weighted towards urban environments and shorter, lower-speed trips.
That’s great, but there are a couple flies in the ointment.
First, obviously, there is no mention whatsoever of non-fatal accidents. These are more difficult to track and categorize, but it seems odd not to include them at all. If the rates of Ubers getting into fender-benders or serious crashes where someone breaks an arm are lower than the national average, as one might expect from the fatality rates, why not say so?
When I asked about this, an Uber spokesperson said that non-fatal crashes are simply not as well defined or tracked, certainly not to the extent fatal crashes are, which makes reporting them consistently difficult. That makes sense, but it still feels like we’re missing an important piece here. Fatal accidents are comparatively rare and the data corpus on non-fatal accidents may provide other insights.
Second, Uber has its own definition of what constitutes an “Uber-related” crash. Naturally enough, this includes whenever a driver is picking up a rider or has a rider in their car. All the miles and crashes mentioned above are either en route to a pickup or during a ride.
But it’s well known that drivers also spend a non-trivial amount of time “deadheading,” or cruising around waiting to be hailed. Exactly how much time is difficult to estimate, as it would differ widely based on time of day, but I don’t think that Uber’s decision to exclude this time is correct. After all, taxi drivers are still on the clock when they are cruising for fares, and Uber drivers must travel to and from destinations, keep moving to get to hot spots, and so on. Driving without a passenger in the car is inarguably a major part of being an Uber driver.
It’s entirely possible that the time spent deadheading isn’t much, and that the accidents that occurred during that time are few in number. But the alternatives are also possible, and I think it’s important for Uber to disclose this data; Cities and riders alike are concerned with the effects of ride-hail services on traffic and such, and the cars don’t simply disappear or stop getting in accidents when they’re not hired.
When I asked Uber about this, a spokesperson said that crash data from trips is “more reliable,” since drivers may not report a crash if they’re not driving someone. That doesn’t seem right either, especially for fatal accidents, which would be reported one way or the other. Furthermore Uber would be able to compare FARS data to its internal metrics of whether a driver involved in a crash was online or not, so the data should be similarly if not identically reliable.
The spokesperson also explained that a driver may be “online” in Uber at a given moment but in fact driving someone around using another rideshare service, like Lyft. If so, and there is an accident, the report would almost certainly go to that other service. That’s understandable, but again it feels like this is a missing piece. At any rate it doesn’t juice the numbers at all, since deadheading miles aren’t included in the totals used above. So “online but not hired” miles will remain a sort of blind spot for now.
You can read the full report here.
In report, Uber discloses 3,045 sexual assaults, 9 murders, and 58 people killed in crashes out of its 1.3B US rides in 2018 — In its first safety report, the ride-hailing company detailed sexual assaults, murders and fatal crashes through its platform. — SAN FRANCISCO …
Uber just released its first-ever safety report that covers sexual assault. Let’s jump right in.
In 2017, Uber received 2,936 reports pertaining to sexual assault, and received 3,045 in 2018. Despite the increase in raw numbers, Uber saw a 16% decrease in the average incident rate, which it suggests may correlate with the company’s increased focus on safety as of late.
Uber categorizes sexual assaults into five subcategories: non-consensual kissing of a non-sexual body part, attempted non-consensual sexual penetration, non-consenual touching of a sexual body part, non-consensual kissing of a sexual body part, and non-consensual sexual penetration.
Regarding the last subcategory, which is rape, Uber received 229 reports of rape in 2017 and 235 reports of rape in 2018. Throughout 2017 and 2018, the reported incidents occurred on 0.00002% of trips, according to Uber.
“While these reports are rare, every report represents an individual who came forward to share an intensely painful experience,” Uber wrote in its report. “Even one report is one too many.”
To be clear, these reported assaults happened to both riders and drivers. Though, Uber found riders account for nearly half of the accused parties across those five most serious sexual assault categories.
“Voluntarily publishing a report that discusses these difficult safety issues is not easy,” Uber Chief Legal Officer Tony West wrote in a blog post. “Most companies don’t talk about issues like sexual violence because doing so risks inviting negative headlines and public criticism. But we feel it’s time for a new approach. As someone who has prosecuted sex crimes and worked on these issues for more than 25 years, I can tell you that a new approach is sorely needed.”
Developing…
Earlier this week, AWS launched DeepComposer, a set of web-based tools for learning about AI to make music and a $99 MIDI keyboard for inputting melodies. That launch created a fair bit of confusion, though, so we sat down with Mike Miller, the director of AWS’s AI Devices group, to talk about where DeepComposer fits into the company’s lineup of AI devices, which includes the DeepLens camera and the DeepRacer AI car, both of which are meant to teach developers about specific AI concepts, too.
The first thing that’s important to remember here is that DeepComposer is a learning tool. It’s not meant for musicians — it’s meant for engineers who want to learn about generative AI. But AWS didn’t help itself by calling this “the world’s first machine learning-enabled musical keyboard for developers.” The keyboard itself, after all, is just a standard, basic MIDI keyboard. There’s no intelligence in it. All of the AI work is happening in the cloud.
“The goal here is to teach generative AI as one of the most interesting trends in machine learning in the last 10 years,” Miller told us. “We specifically told GANs, generative adversarial networks, where there are two networks that are trained together. The reason that’s interesting from our perspective for developers is that it’s very complicated and a lot of the things that developers learn about training machine learning models get jumbled up when you’re training two together.”
With DeepComposer, the developer steps through a process of learning the basics. With the keyboard, you can input a basic melody — but if you don’t have it, you also can use an on-screen keyboard to get started or use a few default melodies (think Ode to Joy). From a practical perspective, the system then goes out and generates a background track for that melody based on a musical style you choose. To keep things simple, the system ignores some values from the keyboard, though, including velocity (just in case you needed more evidence that this is not a keyboard for musicians). But more importantly, developers can then also dig into the actual models the system generated — and even export them to a Jupyter notebook.
For the purpose of DeepComposer, the MIDI data is just another data source to teach developers about GANs and SageMaker, AWS’s machine learning platform that powers DeepComposer behind the scenes.
“The advantage of using MIDI files and basing out training on MIDI is that the representation of the data that goes into the training is in a format that is actually the same representation of data in an image, for example,” explained Miller. “And so it’s actually very applicable and analogous, so as a developer look at that SageMaker notebook and understands the data formatting and how we pass the data in, that’s applicable to other domains as well.”
That’s why the tools expose all of the raw data, too, including loss functions, analytics and the results of the various models as they try to get to an acceptable result, etc. Because this is obviously a tool for generating music, it’ll also expose some of the data about the music, like pitch and empty bars.
“We believe that as developers get into the SageMaker models, they’ll see that, hey, I can apply this to other domains and I can take this and make it my own and see what I can generate,” said Miller.
Having heard the results so far, I think it’s safe to say that DeepComposer won’t produce any hits soon. It seems pretty good at creating a drum track, but bass lines seem a bit erratic. Still, it’s a cool demo of this machine learning technique, even though my guess is that its success will be a bit more limited than DeepRacer, which is a concept that is a bit easier to understand for most since the majority of developers will look at it, think they need to be able to play an instrument to use it, and move on.
Additional reporting by Ron Miller.
Clara Murray / Financial Times : Analysis: mentions of deepfakes or AI-made content in X's Community Notes were more correlated with ...
