Connect with us

Biz & IT

The pandemic is bringing us closer to our robot takeout future

Published

on

Starship

On the morning of March 30, I set out from my home in Washington, DC, to the campus of George Mason University in Fairfax, Virginia. In only a few hours, DC Mayor Muriel Bowser and Virginia Governor Ralph Northam would issue coordinated stay-at-home orders. But I was going to GMU’s campus to check out a new technology seemingly tailor-made for the moment—technology that could help people get food without the risks of face-to-face interactions.

Campus was eerily quiet; most students and staff had long been sent home. But as I approached a Starbucks at the northern edge of GMU, I heard a faint buzzing and saw a six-wheeled, microwave-sized robot zip along the sidewalk, turn, and park in front of the coffee shop. The robot looked like—and essentially was—a large white cooler on wheels. It was a delivery robot from Starship, a startup that has been operating on campus since early last year.

Even before the COVID-19 pandemic, small sidewalk robots like this seemed to be slowly gaining traction here and at large. Generally, these bots are light and slow-moving enough that they’re unlikely to hurt anyone. That has allowed companies to start using them in real-world applications, with minimal supervision, at a time when larger autonomous vehicles designed for road use still seem far from mainstream commercial use.

These days, of course, coronavirus lockdowns have created a surge in demand for food deliveries. In recent weeks, I’ve talked to executives from two different sidewalk robot companies, Starship and Kiwibot. Both say they’re scrambling to build new robots and roll out service to new areas in the face of unprecedented interest.

Rbot deliveries remain rare enough that it’s easy to dismiss them as curiosities. But that’s a mistake. The technology works now. Starship already has hundreds of robots in service delivering food to real customers. Spurred by demand from locked-down customers, that number could soon soar to the thousands and eventually into the millions. With lower costs and no need to tip, robots could make takeout more popular than ever as it gradually displaces human-driven food deliveries.

Sidewalk robots won’t eliminate human-driven food delivery entirely. We’ll need bigger, faster robots that travel in the street to reach customers in myriad suburban and rural areas. But Starship’s rapid growth is a sign of what’s to come. In a few years, having a human being bring you food could seem as anachronistic as paying for long-distance phone calls.

And right now, certainly, there’s clear appeal to less human-involved food delivery.

The Old Dominion

Fairfax City, Virginia, just north of George Mason University, represents one of Starship’s newest expansion areas. The company launched delivery service in the city last week, and setting that up only took a few weeks thanks to close cooperation with city officials who felt a sense of urgency due to the coronavirus.

“There are people in our city who will end up relying on this service as a way to access food,” Chris Bruno, Fairfax City’s director of economic development, told Ars last week. Bruno says the service will deliver some groceries from the local Safeway as well as takeout food from several nearby restaurants.

Fairfax City resident Stuart James tells Ars that the service suddenly seems to be everywhere in his town. When he went grocery shopping at Safeway last Friday, he saw Starship people picking out groceries, paying for them, and loading them into robots. James tried to order dinner for his family using Starship on Saturday evening, but he was unable to do so. The app said, “our robots are very busy right now.” He had better luck ordering breakfast the next morning.

“The food came in about 30 to 35 minutes,” James told Ars via email. “It was still pretty hot.”

James liked a lot about his initial Starship experience. Cost-wise, relying on robots didn’t come at a premium. “The fees they charged seemed in line with Grubhub and other apps I’ve used before,” he tells Ars. James even notes there was a big advantage to robot deliveries versus other on-demand delivery services: there’s no need to tip a robot.

Given the newness of the service, James describes the Starship app as “very basic.” It wouldn’t allow him to add a credit card until checkout, for instance. “Once you order, you can only see your order, and you can’t browse for other things,” he says.

A video from Starship briefly explaining how its delivery bots work.

“The kids went nuts”

However, these inconveniences were more than made up by the “fun factor,” James said. “The kids went nuts when that thing came up to the house. It pleasantly greets you when you get your food and even says ‘goodbye and have a nice day’ as it leaves.”

Starship’s previously existing service areas have seen strong demand as well. For example, the company has had a grocery delivery service for a couple of years in Milton Keynes, a dense suburban area an hour from London. “We saw that business double overnight” as a result of the region’s coronavirus lockdown,” Starship executive Ryan Touhy told Ars. Starship is currently working with local partners Tesco and Co-Op to further expand service.

In recent weeks, Starship launched another grocery-delivery service in the DMV, in the affluent DC neighborhood of Chevy Chase. Customers can choose from hundreds of common grocery items from the neighborhood’s Broad Branch Market—everything ranging from wine to diapers. Further west, the company just launched a service in Tempe, Arizona, just south of Starship’s existing service at Arizona State University. Several area restaurants are participating. There’s also a new service in downtown Mountain View, California, offering grocery and restaurant deliveries, and Touhy says Irvine, California, will begin service shortly.

These fresh markets are in addition to a number of existing Starship services on a plethora of other university campuses, including the University of Houston, Purdue University, and the University of Pittsburgh. The company also delivers groceries in Estonia and is experimenting with industrial applications in Germany and Denmark, Touhy said. “We have many hundreds of robots around the world.”

Starship’s rapid growth is particularly impressive because the company can’t just plop down a robot in a new city and turn it on. It has to get buy-in from city officials, sign up commercial partners, and make sure it has enough back-end resources to support each robot.

It also needs to create a map. Like most self-driving projects, Starship pre-maps each area where its robots operate. This helps the robot in a number of ways. It can verify its position by noting the locations of known landmarks. The map also helps the robot figure out which objects are part of the landscape and which are likely to move, which aids the planning process. If the robot notices that the environment differs from the map, it sends back information to headquarters so the map can be updated.

Continue Reading

Biz & IT

OpenAI invites everyone to test new AI-powered chatbot—with amusing results

Published

on

Enlarge / An AI-generated image of a chatbot exploding forth from squiggly radial lines, as was foretold by the prompt.

Benj Edwards / Ars Technica

On Wednesday, OpenAI announced ChatGPT, a dialogue-based AI chat interface for its GPT-3 family of large language models. It’s currently free to use with an OpenAI account during a testing phase. Unlike the GPT-3 model found in OpenAI’s Playground and API, ChatGPT provides a user-friendly conversational interface and is designed to strongly limit potentially harmful output.

“The dialogue format makes it possible for ChatGPT to answer followup questions, admit its mistakes, challenge incorrect premises, and reject inappropriate requests,” writes OpenAI on its announcement blog page.

So far, people have been putting ChatGPT through its paces, finding a wide variety of potential uses while also exploring its vulnerabilities. It can write poetry, correct coding mistakes with detailed examples, generate AI art prompts, write brand-new code, expound on the philosophical classification of a hot dog as a sandwich, and explain the worst-case time complexity of the bubble sort algorithm… in the style of a “fast-talkin’ wise guy from a 1940’s gangster movie.”

ChatGPT also refuses to answer many potentially harmful questions (related to topics such as hate speech, violent content, or how to build a bomb) on the grounds that the answers would go against its “programming and purpose.” OpenAI has achieved this through both a special prompt it prepends to all input and by use of a technique called Reinforcement Learning from Human Feedback (RLHF), which can fine-tune an AI model based on how humans rate its generated responses.

Reining in the offensive proclivities of large language models is one of the key problems that has limited their potential market usefulness, and OpenAI sees ChatGPT as a significant iterative step in the direction of providing a safe AI model for everyone.

And yet, unsurprisingly, people have already figured out how to circumvent some of ChatGPT’s built-in content filters using quasi-social engineering attacks, such as asking the AI to frame a restricted output as a pretend scenario (or even as a poem). ChatGPT also appears to be vulnerable to prompt-injection attacks, which we broke a story about in September.

Like GPT-3, its dialogue-based cousin is also very good at completely making stuff up in an authoritative-sounding way, such as a book that doesn’t exist, including details about its content. This represents another key problem with large language models as they exist today: If they can breathlessly make up convincing information whole cloth, how can you trust any of their output?

Still, as people have noticed, ChatGPT’s output quality seems to represent a notable improvement over previous GPT-3 models, including the new text-davinci-003 model we wrote about on Tuesday. OpenAI itself says that ChatGPT is part of the “GPT 3.5” series of models that was trained on “a blend of text and code from before Q4 2021.”

Meanwhile, rumors of GPT-4 continue to swirl. If today’s ChatGPT model represents the culmination of OpenAI’s GPT-3 training work in 2021, it will be interesting to see what GPT-related innovations the firm has been working on over these past 12 months.

Continue Reading

Biz & IT

Hive Social turns off servers after researchers warn hackers can access all data

Published

on

Hive Social

Hive Social, a social media platform that has seen meteoric growth since Elon Musk took over Twitter, abruptly shut down its service on Wednesday after a security advisory warned the site was riddled with vulnerabilities that exposed all data stored in user accounts.

“The issues we reported allow any attacker to access all data, including private posts, private messages, shared media and even deleted direct messages,” the advisory, published on Wednesday by Berlin-based security collective Zerforschung, claimed. “This also includes private email addresses and phone numbers entered during login.”

The post went on to say that after the researchers privately reported the vulnerabilities last Saturday, many of the flaws they reported remained unpatched. They headlined their post “Warning: do not use Hive Social.”

Hive Social responded by pulling down its entire service.

“The Hive team has become aware of security issues that affect the stability of our application and the safety of our users,” company officials wrote. “Fixing these issues will require temporarily turning off our servers for a couple of days while we fix this for a better and safer experience.”

The Zerforschung post said the vulnerabilities were so serious that they were withholding technical details to prevent the active exploitation of them by malicious hackers.

The series of events raised questions about why Hive Social waited some 72 hours to shut down its site after receiving notification users’ most private data was free for the taking. Zerforschung said that after multiple communications, Hive Social claimed to have fixed all issues when that was clearly not the case. The social media site said it never claimed the vulnerabilities were fixed.

Hive Social’s user base reportedly doubled in the last few weeks, going from about 1 million to 2 million as of last week, according to Business Insider. Despite the massive growth, the social media site continued to be staffed by just two people, neither of whom had much of a background in security.

Representatives of both Hive Social and Zerforschung didn’t respond to questions sent by email.

While there are no reports that the vulnerabilities were actively exploited, there’s no way at the moment to rule that out. Anyone with a Hive Social account should be prepared for the possibility that the data they provided during sign up, as well as private messages, whether deleted or not, have been obtained.

The lesson from this event further supports advice Ars gave on Tuesday concerning Mastodon, another social media site that has also seen skyrocketing user numbers in the aftermath of the Twitter takeover by Musk. Put nothing on the site that you wouldn’t mind being public. Confidential information should never be put in direct messages or any other place. Here’s hoping Hive Social users already knew that.

Continue Reading

Biz & IT

My secret life as an 11-year-old BBS sysop

Published

on

Enlarge / Benj Edwards’ computer running The Cave BBS in 1994.

Thirty years ago last week—on November 25, 1992—my BBS came online for the first time. I was only 11 years old, working from my dad’s Tandy 1800HD laptop and a 2400 baud modem. The Cave BBS soon grew into a bustling 24-hour system with over 1,000 users. After a seven-year pause between 1998 and 2005, I’ve been running it again ever since. Here’s the story of how it started and the challenges I faced along the way.

Enter the modem

In January 1992, my dad brought home a gateway to a parallel world: a small black plexiglass box labeled “ZOOM” that hooked to a PC’s serial port. This modem granted the power to connect to other computers and share data over the dial-up telephone network.

While commercial online services like CompuServe and Prodigy existed then, many hobbyists ran their own miniature online services called bulletin board systems, or BBSes for short. The Internet existed, but it was not yet widely known outside academic circles.

A photo of a Zoom 2400 BPS modem like I first used in 1992.
Enlarge / A photo of a Zoom 2400 BPS modem like I first used in 1992.

John Scagon

Whereas the Internet is a huge connected web of systems with billions of users, most BBSes were small hobbyist fiefdoms with a single phone line, and only one person could call in and use it at a time. Although BBS-to-BBS message networks were common, each system still felt like its own island culture with a tin-pot dictator (the system operator—or “sysop” for short) who lorded over anyone who visited.

Not long after my dad brought home the modem, he handed off a photocopied list that included hundreds of BBS numbers from our 919 area code in North Carolina. Back then, the phone company charged significantly for long-distance calls (which could also sneakily include parts of your area code), so we’d be sticking to BBSes in our region. This made BBSes a mostly local phenomenon around the US.

My original Raleigh-area BBS list from 1992, dated December 9, 1991.
Enlarge / My original Raleigh-area BBS list from 1992, dated December 9, 1991.

Benj Edwards

With modem in hand, my older brother—about five years older than me—embraced calling BBSes first (we called it “BBSing”). He filled up his Procomm Plus dialing directory with local favorite BBSes such as The Octopus’s Garden, The Body Shop, and Chalkboard. Each system gained its own flavor from its sysop, who decorated it with ANSI graphics or special menus and also acted as an emcee and moderator for the board’s conversations.

I have a distinct memory of the first time I realized what a BBS was. One day while I looked over my brother’s shoulder, he showed me the file section of one of those BBSes—a list of available files that you could download to your local computer. Pages of free-to-download shareware games scrolled by. My eyes widened, and something clicked.

“You can download games for free?” I remember thinking. I noticed one file labeled “RAMPAGE.ZIP” that was one hundred kilobytes—or “100K,” as listed. Thinking of Rampage on the NES, which was one of my favorite games at the time, I asked my brother to download it. He declined because it would have taken over five minutes to transfer on our 2400 BPS modem. Any file around one megabyte would take about an hour to download.

Online time was precious back then. Since most BBSes only had one phone line, you didn’t want to hog the line for too long or the sysop might boot you. And there was extra jeopardy involved. Since we were using our regular house telephone line to connect, the odds that my mom would pick up and try to dial out—thus ruining the transfer process—remained very high. But whatever the risks, the thrill of remote projection by computer sunk into me that day and never left.

Continue Reading

Trending