gpt2_articles.txt

=== GENERATED SEQUENCE 1 ===
TITLE= Amazon to Retrain a Third of Its U.S. Workers as Automation Advances TEXT= So far, the American workforce has been largely in workers of color. But Amazon is about to ramp up hiring of some of its more 200,000 hourly workers, according to a person familiar with the company’s plans. The new hires won’t be those in the same field as its more self-professed “white-collar workers,” as the company abbreviates its growing staff. Instead, the roughly 1,000 new workers will be drawn from two large and multiracial groups: Amazon has laid off thousands of Black and Latinx workers, has laid off dozens of Latinx employees, and is currently seeking Federal Employment Development money to pay for new workers.“Right now, the numbers of Black, Latinx, and African American workers are shrinking, because there’s still not enough talent available to continue these high-volume, high-pressure work,” the person familiar with Amazon’s plans explained. Amazon declined to comment on the more than 1,000 new positions it plans to employ.Many of those workers are middle-class white men who choose to build their careers on
=== GENERATED SEQUENCE 2 ===
TITLE= AI researchers propose making chatbots and voice assistants more like their real-world peers TEXT= Chatbots, which are a form of virtual personas, have long been the interface between human and machine. Advances in the past few years have democratized voice assistants, responding to the needs of multiple speakers at once, aiming to reduce set times and improve duration. But much of the field has proven difficult to apply conversational AI effectively in the real world, especially on sensitive tasks involving such as sales or customer service.To address this issue, a team from MIT Technology Review (MIT Tech Review) and Beijing-based research institute Beijing Brain Institute have developed a novel approach to training conversational AI systems on a dataset of short conversations over long periods of time. The researchers developed a chatbot named QUEBO after their lab name, which stands for QUEBO, "Quō-" or "QUE-BO." They say the research will serve as a benchmark to benchmark conversational AI’s capability to handle small conveniences like social media conversations in real time
=== GENERATED SEQUENCE 3 ===
TITLE= A fight for the soul of machine learning TEXT= Earlier this month, researchers at Nvidia, one of the biggest artificial intelligence (AI) companies, revealed an unwieldy and potentially game-changing deep learning algorithm called "neural pre-training." In essence, it was so pretrained that it learned far better by trial and error than a human who had been explicitly trained on the same data set.The results promised an entirely new way to train AI. In doing so, it essentially taught itself to do things no robots had ever seen.The result, known as a neural pre-training neural network, or neural network ensembles, has become a favorite of machine learning researchers for its incredible abilities to make breakthroughs in just a few hours. The concept is surprising not only for its potential power and promise, but also because it represents a major paradigm shift in computing. From manually designing and training data sets to finding optimal solutions to streamlining AI projects, pre-training has given rise to a cottage industry dominated by startups and big companies intent on streamlining and streamlining their AI projects.The promise of this new technique, called "deep learning on steroids," originated at Google and proves highly effective at increasing performance across a range of machine learning
=== GENERATED SEQUENCE 4 ===
TITLE= Spacemaker Wants to Make Your House Feel Like a 'City of Scapes' TEXT= Since Elon Musk took control of the spaceship roving Mars in 2012, hundreds of house robots have taken to the road. But not all scuttling efforts will be on schedule — the nation’s first Spacemaker, which is being constructed in Cape Canaveral beach near Florida.The machine is the brainchild of former Tesla CEO Elon Musk, who is best known for his Model 3 trucks, which he has dubbed the Model 6. Musk said the first Spacemaker will be operational by mid-2023. And though many expect the Model 3 to be an enormous undertaking, it will make progress — because unlike previous motorized vehicles, things can change.Musk has said the first Spacemaker will be human-powered by 2020. But the computer will still be hooked up to the grid, meaning it will rely on local power. That will make it tricky to power the vehicle — unless the geoffices are set correctly — and leaves it vulnerable to battery shortfalls.Tesla is known for making consumer-grade cars, but for long-haul delivery, dealing with unpredictable conditions or power failure is its highest priority.Last year, Musk promised that Tesla
=== GENERATED SEQUENCE 5 ===
TITLE= Alibaba Cloud publishes machine learning research papers, research papers, and talks TEXT= A number of research and development companies today announced a range of research-related announcements. Listed below are some of the titles from the paper ALBERT: Natural Language Processing Datasets for Deep Learning Supercomputing.Paper: Neural Decisions via Generative Adversarial Networks Generated via Machine Learning, by Shangzhe Wu, Daniel Balouek-Thomert, Sekhar Tatikonda, Diego Melgar, Xiaolin Ding, Shaoqing Ren AbstractThe demographic composition of the detected models in this study reflects the need for improved methods for addressing problematic dataset subpopulations. We further refine the task of classifying large and diverse (>50M) neural networks by jointly estimating the family of family of family trees, focusing on subsets of the family I, II, or III. We also propose a more coarse-grained view of model families by reconstructing feature extraction and coarse-grained elevation. Our experiments show that, through appropriate modeling strategies, these
=== GENERATED SEQUENCE 6 ===
TITLE= Researchers develop a traffic light for self-driving cars TEXT= A pair of new papers from University of Tokyo Institute of Technology researchers and the World Health Organization claim to have addressed one of the great challenges of artificial intelligence: traffic lights. Thanks to machine learning, they can illuminate traffic for much more effective and safer operation than they could on paper.In research authored by Atsushi Sakamoto and colleagues at the Institute of Intelligent Systems and Decision Support in Tokyo, the team presents an efficient way to turn traffic lights on and off. In practice, however, identifying and setting the timing for lighting within a traffic space can prove difficult, as it’s often unpredictable and there may even be visual clues that indicate the start of a car ahead or the end of a short one.Sakamoto and colleagues came up with an efficiency-enabling solution based on approaches developed by researchers at Google, IBM, and Toyota. The team’s approach isn’t meant to be, nor indeed is it currently available, but it provides an easy-to-use and
=== GENERATED SEQUENCE 7 ===
TITLE= Amazon to Retrain a Third of Its U.S. Workers as Automation Advances TEXT= So far, American workers have made up just 18.6% of the net employment that the United States produces in employment, according to the latest estimates from the National Employment Forum, a coalition of business leaders. This represents less than 1% of the total U.S. workforce and less than 4% of net employment. While the federal government provides some relief, it covers much more than this. Together, American workers produce more than 60% of the net employment, amounting to more than $13 trillion per year. To make up that missing piece, companies in industries that were expected to add jobs to workers and then go bankrupt are struggling to hire as many workers as they need to, says the coalition report.The jobs lost and gained in this way are not easily quantifiable. “ Automation has a very long history of being a tailwind to productivity growth,” says Erik Brynjolfsson, CEO of Brynjolfsson & Co. “The reason for this tailwind is the mismatch between how workers are represented and how production moves into those new roles.”The federal government incentivizes companies to
=== GENERATED SEQUENCE 8 ===
TITLE= Spectrum Labs raises $45 million to bring you next-gen entertainment TEXT= Everything jobless Americans need to know about the $45 million InHub One round, co-founded by venture capital firm OpenAI, is about to see a tsunami of excitement as companies test the waters of next-gen technologies such as AI and virtual assistants.With more than 1,200 employees globally recruited, InHub One will bring next-gen talent from companies including Netflix, Spotify, Google, Amazon, and Netflix parent company Yandex. It’s part of the company’s broader strategy to focus on creating an ever-more-capable talent pipeline — and it’s not just bringing next-gen talent to the lab. InHub One, which is a part of the company’s parent company Alphabet, said Thursday it’s working with 600 Netflix executives to ink next-gen deals with emerging companies, among others.TITLE= AI Camera Ruins Soccer Game After Mistaking Referee's Bald Head For Ball TEXT= Technology and sports have a fairly mixed relationship already. Log on to Twitter during a soccer match and someone might see a bunch of artificial intelligence-powered cameras tracking the opposition and attempting to foul the ball.
=== GENERATED SEQUENCE 9 ===
TITLE= Augmenting artificial intelligence with bias can help solve problems in public housing TEXT= Augmenting artificial intelligence with bias can help solving problems in public housing, according to research led by researchers at MIT and the University of Chicago.Part of the challenge in managing the spread of disinformation is ensuring that candidates are aware of the bots’ potential biases and of the fact that they might affect public policy. It is part of a coordinated effort by academics and companies to zero in on ways to neutralize the bots, dispelling any concerns about their casting doubt on the efficacy and effectiveness of their online advertising. The bots live online and in the real world, collecting and analyzing large amounts of information on a given user’s intentions and actions.Because bots are such complex and pervasive, studies are necessary to understand their persuasive power and, in some cases, to dispel the bots' influence, according to Noam Brown, an assistant professor at the university and coauthor of a new paper detailing those efforts. That requires new insights into the ways in which artificial intelligence can be leveraged to counter the influence of bots. That’s partly what Humayun Noam, an MIT PhD student, wanted to begin by looking at ways to automate the assessment of
=== GENERATED SEQUENCE 10 ===
TITLE= Artificial Intelligence Advances in Health Care Will Be Slow, Not Slow TEXT= Artificial intelligence (AI) is widely in use today; some of the most exciting demonstrations of it come from applications such as conversing with your family or robots playing a passive role in a virtual world, or inverse reinforcement learning, in which a machine replaces an individual’s decision-making process in a virtual world to recommend something to do online. Other forms of AI, such as search and social media recommendations, use pattern recognition to target ads for specific groups of people. However, these technologies still require a lot of computation to perform well—and current research suggests that the rate of improvement will only get better if we can accelerate the use of AI to improve healthcare outcomes.Artificial intelligence (AI) is now widely employed in the healthcare industry, used extensively to improve or even ensure the performance of medical diagnostic tools. Researchers make significant progress in the last decade or so, building on stunning advances in basic AI fundamentals—key computational principles that allow machines to reason about many complex scenes and situations—and applying these principles to medicine. These principles lead to better, more accurate and effective clinical care, advancing diagnostic technologies like X-rays or time course predictions for a person undergoing cancer
=== GENERATED SEQUENCE 11 ===
TITLE= You can get a robot to keep your lonely grandparents company. That robot is surprisingly expensive.Research shows how AI can make companies more financially sustainable TEXT= Share All sharing options for: You can buy a robot to keep your lonely grandparents company. That robot is surprisingly expensive.A University of California, Berkeley research team has developed a computer program that’s capable of taking a person out of bed for eight hours at night and automating it with video and voice input.“Companies are trying to make people not want to come out of their rooms at night,” says Gary Marcus, UCLA assistant professor of media arts. The company is named after its lovable character Liku, from the Pixar film that’s more affectionately dubbed Liku the Friendly Ghost.The company is making its technology commercially available, and it’s not the first to make use of the tech. Researchers have been using virtual assistants to let lonely people get extra company—or, in this case, company shares. The Liku system is at least as advanced as a traditional Bluetooth headset, says Andrew Huang, UCLA assistant professor.Companies may be tempted to let the employees use their real-time phones to text or call their boss when they need some
=== GENERATED SEQUENCE 12 ===
TITLE= The Military Wants Better Tests for PTSD TEXT= Many people are still struggling with posttraumatic stress disorder (PTSD), a common anxiety disorder caused by high levels of stress during war. Some psychiatrists use a computer to assess patients, trying to determine how they might respond to everyday circumstances, like guns or knives, or talking on the phone. But the Air Force is hoping to change that by enlisting the help of companies like Clarifai, a company that’s trained thousands of psychiatrists to use its technology.A recent survey conducted by the U.S. Army through its war-fighting research arm found that only about half of respondents thought that their company’s products would be effective at predicting the effects of PTSD symptoms after five years of follow-up interviews. Other research suggests that only about a quarter thought that any kind of new technology would be effective. So the Air Force is taking steps to trial the technologies in hopes of helping its troops adapt.Clarae Robertson, the company’s lead researcher, says the trial is the first test of its kind among modern war-fighting operations. “We wanted to see if we could have the same impact on the way people cope with post-traumatic stress disorder,
=== GENERATED SEQUENCE 13 ===
TITLE= Amazon to Retrain a Third of Its U.S. Workers as Automation Advances TEXT= The world’s economy continues to expand, and in the millions millions millions of workers are losing their jobs to automation as smartphones, virtual assistants, and home appliances add to their already burdensome workforces. Amazon is retraining a growing workforce of more than 1 million workers, as companies accelerate moves toward mass-scale automation.The American workforce is expected to grow to more than 2.6 million by 2025 from less than 10 million in 2018. And for the first time, companies are hiring more skilled welders and computer programmers to help their companies automate more advanced manufacturing tasks — a recent turn in the company’s industrial strategy.Automation is about to create a third of U.S. jobs, and companies are hiring more and more workers. As companies accelerate the rollouts of new automation strategies, they will do so in ways that further add to economic growth.For starters, Amazon, Target, AT&T, and Walmart have announced their initial public offering incentives to help workers automate more advanced manufacturing. Employers receive cash and stock in the companies’ warehouses, enabling them to invest in the workers’ needs.“
=== GENERATED SEQUENCE 14 ===
TITLE= Patterns of Artificial Intelligence Are Biased. Here's How They Are Being Made TEXT= Over the last two decades, humanity faced two major challenges. Forerunners of the field of Natural Language Processing (NLP) developed techniques for identifying patterns in written text. Advances in artificial intelligence (AI) have enabled researchers to apply these techniques to speech or images, making them essentially mimetic tools of the human brain. But while these forms of AI outperform human vision, they also raise serious concerns about their accuracy, efficiency, and privacy.In this research, we review the recent developments in artificial intelligence (AI) and the ethics of using these technologies, and argue how the technologies are being shaped by and contributed to these third fields of research. We also discuss recent efforts to regulate AI and the human condition and offer a perspective on the implications of these technologies.Artificial intelligence is a general-purpose enabling technology with applications anywhere within the economy. For example, it is used to automate aspects of decision-making and supports the routine collection and management of personal data. It is used in games, pattern recognition and to monitor people’s emotions and whereabouts. It is being deployed across borders to measure economic performance and to detect hate speech. And increasingly,
=== GENERATED SEQUENCE 15 ===
TITLE= State of AI Report 2019 TEXT= State of AI Report 2020The State of AI Report is the most comprehensive look into the State of AI Report and provides insights across departments, research groups, industry, and business leaders. You can find our full report here.The AI Index covers the following topics:Applying AI Research To Natural Language Processing (NLP) ResearchResearchers at Google, Facebook, DeepMind, and academia have laid the foundations for the State of AI Report 2020. In this Report, we look at how they are using AI for natural language processing (NLP). This is a quest that is still in its infancy, but has rapidly accelerated in the past five years.Here are our key findings:Despite the progress, there is still much to do before we can apply NLP research to fundamental AI tasks such as language translation, language modelling, or autonomous driving. Such work will take many years — in many cases a decade — before truly “general” AI can be applied to a wide range of tasks.To start with, we identify three areas where we think NLP research can make a difference:Methodological shift Methodological shift We see major potential in using more rigorous methods early in the AI Index process, particularly when it
=== GENERATED SEQUENCE 16 ===
TITLE= At Starsky Robotics, we're unlocking the next generation of robots with machine learning TEXT= At Starsky Robotics, we believe that the future of work can be defined by collaboration between humans and machines. This can best be understood by mapping the physical world around a workstation and employing computers to execute a task. The machines are initially designed with wheels to move about, others to stay in the frame, and a camera system in the circuit for image recognition and manipulation.At Starsky we develop our machine learning (ML) platform, called Tensor Processing Units, to drive the computation that actually makes our workplaces and products perform. The idea here is to move from modeling to decision-making and from intuitive manipulation of large-scale data to decision-making in the real world. It is designed to solve many of the challenges that are inherent to jobs in the manufacturing, warehousing, health care, education, and elsewhere and to empower workers to achieve a purpose that isn’t achieved just by making technology cheaper. By enabling firms to operate more cheaply, we can position them to compete in an increasingly globalized economy.At Starsky, we’re making progress on just such a future. In 2019 we announced our strategic partnership with Carnegie Mellon
=== GENERATED SEQUENCE 17 ===
TITLE= Artificial Intelligence Hits the Barrier of Meaning TEXT= Reading this article has opened my eyes to the many challenges that come with using A.I. The first is simply breathtaking in scope and scale. But there is also a saying that if we can find a way to use this amazing technology for so many important purposes, we will all have an easier life.In this article I want to share with you a simple and very important truth: without the use of A.I. the gulf between what real technology can do and what you would find inside an A.I. box. A.I. that exists beyond all technology is the future of humanity.Promotion and commercializationIn a beginning industry dreams come true: there is a day when genuine breakthroughs will be achieved by bringing the use of A.I. into the mainstream conversation. An A.I.-generated breakthrough will not be the product of some lone genius, but a clever team of talented researchers.Within the past year there has been a lot of discussion about the future of artificial intelligence (A.I.) and the implications it may have for society. To answer this question, we need to look beyond the hype and speculations surrounding breakthroughs and truly understand the actual capabilities of A
=== GENERATED SEQUENCE 18 ===
TITLE= Best week ever for autonomous driving TEXT= Hot on the heels of Tesla's public debut autonomous car event, Waymo is taking the wraps off another wildly ambitious program with a few new segments. This week, the two companies share some exciting developments in a new series of interviews.1. How autonomous car tech can scale up Waymo’s operationsImage: WaymoIn the early days of their self-driving car program, the self-driving unit of Alphabet focused on building a tool that goes a long way toward automating driving. Waymo wanted to build a tool that goes well beyond driving, allowing cars to monitor the environment, look for dangerous situations, and take actions as required.But with test deliveries set to take place over the next several weeks, Waymo realized it couldn’t scale that up. So it created what it calls a “driver assist” feature.This driver assist system is based on a neural network that Waymo’s cars use to check in on their surroundings, looking for things like a vehicle on the road ahead or a bright light in front of them.Human drivers drive based on a training set, with test points set up along the way. For example, the bot can detect traffic
=== GENERATED SEQUENCE 19 ===
TITLE= Research summary: Troubling Trends in Machine Learning Scholarship TEXT= Mini summary (scroll down for full summary):With the explosion of people working in the domain of ML and the prevalence of preprint and peer review in ML conferences like arXiv and arXiv, there are some troubling trends regarding the scholarship of ML.With a flood of papers with tons of promising papers, many have started to feel that they are “unbalanced” by authors. I am talking about some authors who have been working in the domain of ML for a long time and have seen a flood of papers that are seemingly good at providing some insight into areas that are not quite right. These include questions like “What is the strongest contribution of the strongest paper?” and “What is the most surprising research direction?”The current way of thinking about research direction is quite vague and leaves many with no idea what they are doing. One common misconception is that the best research direction is usually driven by the authors. In reality, there are many factors that play into the direction of a research and there are many different directions that a research has been going and many different directions that have not been explored in detail by all researchers.This flood of papers is
=== GENERATED SEQUENCE 20 ===
TITLE= DeepMind’s Dreamer AI learns from past videogames to generate dreamlike phrases TEXT= Some of the most striking recent examples from artificial intelligence come from tracking gamers by tracking their movement. That’s a useful way to measure a person’s “attention,” but Dreamer AI’s system begins by simply collecting a phrase and then generating sentences in natural language to track a player’s whereabouts, as a pattern. That’s surprisingly natural and close to capturing how gamers form associations, and suggests a new method of tackling a problem that’s both challenging and intellectually stimulating.Such repetitions are particularly resonable when they match a goal, such as tracking a bird across the field. Alternatively, if you’re stuck, say, tracking an enemy, the Dreamer system generates sentences in natural language to match you. Such sentences become connectional within Dreamer, responding to the context of the sequence of words by generating corresponding sentences, even when they don’t match the context.This approach
=== GENERATED SEQUENCE 21 ===
TITLE= Nvidia unveils Maxine, a smarter AI chip with more processing power and much less latency TEXT= Nvidia today announced a new $10 million round led by Greycroft and Bee Partners, with participation by 2Enable, a maker of software to aid engineers with disabilities.Grady Booch, CEO and founder of Greycroft, said the funds will help firms accelerate their AI R&D, explore AI applications in new AI-related projects, and put the technology to use when it’s ready. Funded by a $10 million Series A round from Greycroft, the funds will accelerate work to develop AI that delivers benefit-value equitably, quickly and broadly, increasing the net impact of AI on employee health, environment and workforce safety.“It’s amazing to see a two-year-old startup with $200 million in VC backing deliver meaningful, practical AI that’s less biased,” Booch said. “To have this backing deliver value for somebody’s immediate financial advantage is unheard of.”
=== GENERATED SEQUENCE 22 ===
TITLE= State of AI Report 2019 TEXT= State of AI Report 2020The State of AI Report analyses the progress of artificial intelligence (AI) in several key areas. The Report considers the following key dimensions: Research: Digital transformation; Human-computer interaction (including AI) and domains of AI; Technology-driven improvement; Labor market and economic growth; AI safety, governance, and ethics; and Predatory and emerging AI threats.The first 10 areas identify three types of AI: research, deployment, and adoption. The adoption of AI will likely cause economic growth, productivity, and accessibility of routine business functions, increase productivity and employment, and improve the performance of central and local AI systems. The adoption of AI will likely cause falling unemployment, but there will be significant social disruption as AI becomes a high-technology, high-technology or service technology. The sector best positioned to impact the adoption of AI in economic and policy domains will be the “highly-developed” (e.g., automotive) sector — AI that is mature enough to create productive new jobs and predictable enough to allow government agents to effectively manage the massive following:Production-ready AI: Will enable machines to perform complex, real-world tasks; Assist AI devices and organizations in enabling
=== GENERATED SEQUENCE 23 ===
TITLE= Reinforcement learning can boost creativity and debilitarize workers TEXT= The market for reinforcement learning algorithms has taken a big downturn, says Gary Marcus, the director of research at the robotics firm Fractl and a member of the board that hosted the meeting.The firm, which was founded in 2013, expects to see revenue fall by the end of 2019 from a combination of a continued decline in core software and hardware sales, and falling advertising revenues. Its last great revenue came from on-selling products, including products popular with some roboticists that have seen a resurgence in their sales following recent AI-enabled "re-branding" sessions.Also: How to train your business AI (free PDF)The biggest reason to discipline robotics is fear of failure. With so much progress under way, reinforcement learning appears to be having an effect, with companies turning to increasingly difficult forms of AI such as natural language processing, reinforcement learning, and imitation learning to increase their revenue streams. Reinforcement learning has been around for years but its promise has been extended further when "deep" robotics, computer games like Pong, and robotics running continuously, have become possible.That's no guarantee that the market will again, or that Fractl can raise enough big bucks to
=== GENERATED SEQUENCE 24 ===
TITLE= Waymo's robotaxi pilot surpassed 400 in just one day TEXT= Waymo transported 6,678 passengers in its first autonomous taxi ride in Phoenix, Arizona on Sunday, bringing the total number of passengers in a commercial ride network to 8,678.AdvertisementThe driverless Waymo vehicles worked around 5,000 rounds per hour, the company reported. Most passenger cars had been pressed for pickup and were told to wait outside for about 45 minutes, according to the teletexted minutes the company reported. No passengers were hurt.In total, Waymo transported 6,678 passengers in its first commercial ride-hailing service in Phoenix, according to Waymo CEO Kevin Johnson. Despite some delays, his robotaxi pilot did eventually reach a high speed of about 45 minutes.Waymo didn’t disclose which other companies or routes were used to test the autonomous technology, but Steve Mahan, director of automated vehicles for the Phoenix-based ride-hailing startup, said in a statement it had test-pilot flights with autonomous vehicles on standby while he reviews the pace of passenger movement. The company did not share the total number of rides Waymo is currently using, though it did reveal that its robotaxi pilot had already reached
=== GENERATED SEQUENCE 25 ===
TITLE= What is radiology AI? | GalleryIT revolutionizes diagnostic radiology imaging by allowing a machine to not only understand but understand all of the visible signs of disease in a matter of hours. The potential of this technology is enormous.At a time when the diagnostic radiology workflow is in its infancy and routine is changing, the potential of radiology AI is becoming ever more critical. AI will soon become the diagnostic AI of the future.At the same time, the availability of diagnostic images is only just beginning to make radiology the industry’s most sophisticated set of tools. AI will improve the quality of disease assessment, detect times and obtain personal recommendations on treatment. This will pave the way for an ever-expanding list of imaging technologies that will be used, by and for everyone, to quantify, classify and screen a population of already underserved patients.We cannot wait for the days to come when this exciting technology will be deployed at the cutting edge of medical imaging — and at the right, to the patients and their caregivers.Don’t forget to follow us on Twitter to stay up to date on the latest news and research in the world of radiology and AI.TITLE= AI could diagnose cancer early by analyzing chest X
=== GENERATED SEQUENCE 26 ===
TITLE= Activists Turn Facial Recognition Into a Crime-fighting Tool TEXT= Some activists believe that a technology that’s already too powerful can be used to turn criminal suspects into violent criminals by turning facial recognition against them.They’ve dubbed facial recognition the industry’s “lethal solution” in an article published this week. In it, sociologist Joy Buolamwini and Timnit Gebru, then at Microsoft and Stanford University, argue that facial recognition, which they call “video surveillance” technology, is not just bad but dangerous — because it misidentifies people and erases their faces from the screen without their consent.Before they introduce the “lethal” tech to police, many cities across the country have started to implement them. Some use algorithms that identify who is most likely to commit a crime, like mug shots or license plate images, or a crime that involves arrests and warrants.But there’s plenty of debate about whether video surveillance can actually reduce crime. Some researchers say it can help police find perpetrators and victims before it zooms in. A few months ago, researchers at the Massachusetts Institute of Technology found that facial recognition had a strong negative impact on communities of color
=== GENERATED SEQUENCE 27 ===
TITLE= How I used NLP's GPT-3 to write the AI I Created TEXT= When it comes to creating AI, there are 2 main choices I have:- write a clean code or- just use the output from the original app to clean code.Either way, there are a lot of good articles and code examples on arxiv.com.The reason I choose the latter is both lovely and surprising: with a solid foundation of 30 lines of code (more on that later), this way of creating and using AI is fairly simple. I wanted a clean, sharp and concise writing experience that would let my coworkers and I take on difficult creative problems with clear writing.What I didn't expect was for the first part of the code to become a bottleneck, exposing the messiness of the AI. The second reason may be more complex: I was writing this sentence while driving, and as I was about to cross the street, I accidentally sent the car driving by on the bright day and ended up writing another 🤔code 🤔 to keep driving.Despite its simplicity, the problem with GPT-3 is its inability to generate long sequences. If I cut out a long word (‘a’ is good,
=== GENERATED SEQUENCE 28 ===
TITLE= Watch Out! Google DeepMind & Facebook AI Just Made Algorithms That Can Pass For Turing Tests TEXT= Just as the artificial intelligence breakthroughs of the past can be described with a few sentences, the latest developments in machine learning can be described with greater detail.These algorithms have driven many of the most remarkable advances in AI: They enable computers to write songs, beat the world champions at board games, tell fake stories, generate language, and even make predictions. But they have fallen short in other domains, like answering factual questions (are you a Christian or an Arab) or providing recommendations to a politician (is the music in your favorite band good or hard?).Now Google and Facebook are promising to change that, with a new deep learning algorithm that can pass for human test interpretations even without extensive knowledge of the subject. Called ADEPT (Adversarial As-Learning To-Answer-the-Question-and-Conversation-Advanced) by researchers at Google’s DeepMind and Facebook AI Research, the new software stacks these cognitive approaches to an existing task with classical strengths of natural language processing, or NLP.“It's the opposite of what deep learning was doing for answering the natural language problem,”
=== GENERATED SEQUENCE 29 ===
TITLE= Towards a more AI-enabled Argentina TEXT= Today, as the world confronts the outbreak of coronavirus, I decided to see what could be done to increase the number of AI researchers in Argentina. So I spent over a month working with researchers from around the world to put together this list of some of the most promising places to teach AIs to learn about machine learning and neural networks.Establishing a presence in the AI community? Buenos Aires must succeed in making the environment more hospitable to AI.While some countries such as Brazil, Ecuador, and India have embraced AI, Argentina is far from that place. There are comparatively few people applying their work in machine learning and there is very little that makes a big difference to where someone goes – I spoke to friends and colleagues in Argentina from many different backgrounds about where to go next.While many expect that advances in AI will trickle down to universities and companies, there is tremendous power in numbers to make these opportunities more accessible.In Argentina, for example, there are over 120 colleges that are tasked with supervising students, teaching them programming skills, and providing quality care and support to the elderly.I went to Argentina to meet with a number of top researchers to discuss where Argentina could go next.
=== GENERATED SEQUENCE 30 ===
TITLE= Robot farming startup Iron Ox has emerged from stealth with a self-driving prototype TEXT= The Silicon Valley robotics company has made headlines this week for attempting to clean a field in Australia using 1,000 pounds of chemical. But it’s not the first time the local robot company has faced a local backlash. Last month, for example, Uber and Lyft were hit with fines of up to $5,000 for using the chemical in their fleets.Iron Ox is one of a number of companies trying to automate repetitive agricultural tasks using self-driving or robotic systems. Robots are used to transport heavy equipment such as drills and pick-ups. But these machines are expensive and cannot be deployed without a human on board. The company says it has had to disable a robotic assistant which takes control via a phone app, but continues to make money from the unit.Machine needs an artificial intelligence environment to gather dataThe iron ore is dug into a shallow hole, where it is heated by 1,000 pounds of hydrochloric acid (H2CO2), using a heated electrode tip. The solution is then heated to a temperature of 46,000 degrees. This creates a liquid that is the building block of artificial intelligence, according to Iron Ox website.Iron
=== GENERATED SEQUENCE 31 ===
TITLE= Google Gives Its Human-Like Phone Chatbot A Demo Redo TEXT= Some of Google’s most buzzed-about mobile products today include the Pixel phone and the Duplex internet-connected speaker AR. But the company’s latest research (demo) highlights something a bit less sexy than the pitch deck.Google is enlisting the help of tech giant H20 to run its own "human in the loop" audio reinforcement learning (RL) machine learning experiments as an early proof of concept. Google says it’s done the “tremendous job” of reaching a public beta of its Duplex virtual assistant, which it calls Duplex 2.0, and wants to give it a first bite. The startup has two goals:To demonstrate how the system “learns” without the help of humans and without reinforcement learning — in a fun way that is also largely unproven given existing tools that learn continuously, making learning challenging — and to lay out a foundation for humans to manually input input their input so
=== GENERATED SEQUENCE 32 ===
TITLE= Privacy Advocates Herald Bill to Curb Corporate 'Privacy Hacking' as it Continues in Utah TEXT= The American Civil Liberties Union on Wednesday launched a nationwide campaign pressuring the Utah Department of Commerce not to add new protections to whistleblower protections that have come under increased scrutiny after the companies that collect records on citizen privacy and use the data on citizens to push them to comply with company policies.Among the companies now under boycott by states and localities include Access Now, Verizon, Unilever, and Apple. Also under boycott is American Express.As The Daily Beast previously reported, over the last few weeks, employees at some of Utah’s largest companies have been warned not to tell anyone their data is being used by other companies or organizations unless compelled by state or federal privacy laws. In response, public records requests and litigation by groups like the Electronic Frontier Foundation, Amnesty International, and Common Cause have demanded that the state stop police use of Clearview AI’s technology, face recognition, and other tracking technologies.In response, the companies have said they won’t sell data collected by the company to outside researchers or Illinois-based data brokers, and that they won’t sell access to the data to outside researchers or Illinois-
=== GENERATED SEQUENCE 33 ===
TITLE= Meet Microsoft's first AI-powered coffee machine TEXT= Jeff Dean, chair of Microsoft’s artificial intelligence (AI) division, recently sat down with WIRED senior writer Will Knight to discuss the value of building products that do a lot of good, but don’t always do much of anything else.Dean tells WIRED that he thinks we’ll all need to find some sort of “general purpose product” — a conceptual mashup of software and hardware that completely reimagines the capabilities of a specific computer (or, more generally, a business) in order to tackle one task. It’ll probably require buy-in from product builders, analyst agnostics, human resources departments, call centers, product managers, and a few other well-intentioned minds, but Microsoft has a pretty good shot at that right now.Dean tells WIRED that while there’s certainly value in building product companies that “do something interesting,” they also need to do it in a way that can tackle customer needs. There’s a certain inevitability that when companies start leaning into AI (or deep learning), they’ll need to think big.“Deep learning is not going
=== GENERATED SEQUENCE 34 ===
TITLE= Robots could help save your county from the flu TEXT= The iconic clockworkwork clock is falling apart. As people scramble to re-start life-saving work around the globe, scientists believe artificial intelligence could help keep it all going in a cooler, more human way.At least, that's the theory being proposed by researchers at Stanford, designed to lay out the challenges ahead. Artificial intelligence (AI) and machine learning (ML) could bring about an entirely new crop of jobs, the experts said.In research published in the scientific journal Nature Machine Intelligence, researchers reveal how current methods can often take AI months or even years to accomplish. And given the current pandemic, it's unlikely a task like remote work will be far behind.Rather than re-start the clock, scientists hypothesize we could instead use machine learning to re-create the essential elements of human lives – the ability to work during the pandemic, for example, and the ability to balance work and social life during a pandemic."If we can use common sense, how do we get the things we need to do to stay healthy?" study author Ioannis Costanzaiu, assistant professor of computer science and center fellow, said.For some people, the idea of automation
=== GENERATED SEQUENCE 35 ===
TITLE= The U.S. Is Too Hot to Handle the Elastic Compressive AI Revolution TEXT= From 2012 to 2018, the world's top artificial intelligence researchers — AI pioneers DeepMind and Google — generated a whirlwind of press coverage. One widely covered story said the “AI revolution” was about to see an explosion of new machine learning research — some big and fast. Another said the industry was about to enter a plateau and require policy changes. In 2014, the Wall Street Journal called the AI boom “the biggest AI success story of all time.”There was also a lot of public hype. But what really took off in 2012 was the emergence of the “AI superpowers” — multibillion-dollar AI companies with powerful AIs that surpassed human-level expertise in a field known as artificial general intelligence (AGI). With about 100 billion parameters, or AI methods — huge datasets of data sets containing something like 6 trillion parameters — deep learning powers supercomputers with superhuman ability to tackle human tasks like speech and driving.Many people thought this would be a great time to upgrade AI, so they invested. But the field remained sluggish, and so hot was the AI revolution that some researchers and investors wanted to turn the
=== GENERATED SEQUENCE 36 ===
TITLE= The Birthplace of AI TEXT= The Birthplace of AII. THE BENEATH OF AIIn 1945, at age 22, my father, the inventor of the atomic blowup, was diagnosed with cancer. The following year, after my second diagnosis, I was diagnosed with breast cancer.This diagnosis was not unexpected. For two very different reasons: I had been diagnosed with cancer before (i.e., my father was diagnosed with the disease at the same time) and after; and two other diagnoses seemed entirely plausible, since I was diagnosed with both cancers at the same time.Even though baseball is physics, physics is not what matters when you want to predict the future. Two different approaches to cancer were being tried — one by machine and another by nature.The machine approach involved two opposing approaches. For baseball, the doctors measure base hits on a batter's head, then measure the hits on a second baseman's head during each base phase. Based on these two approaches, the pitch is compared pitch by pitch. The artificial approach looks at baseball's 3-point trajectory, the hit batters head by head, and the two ways the pitch is used change how the season goes.My father looked at both approaches, first trying both first and
=== GENERATED SEQUENCE 37 ===
TITLE= Artist's conception of AI-generated art TEXT= AI-generated art can be pretty wild and weird, but this AI artist has found a way to make himself known.Since his San Francisco-based agency, Obvious, began drip-feeding its AI artwork last October, the public has been using the colorful, never-before-published images to share and discuss his latest work.This isn’t the first time Obvious has taken its unusual approach to art, says Cillian Debnamid, senior program director of digital media at the creative agency. Obvious began with traditional sketches and sketches, then artists paint over new ones.“It’s kind of like finding a theme,” Debnamid says. “You kind of grab the ideas and run with them until you get something that works.”One of the side effects of using AI is that people can see the style and parameters used to create the artwork. Sometimes it’s an approximation of what a human artist would do, using still images and lines from work that was produced by an AI. Other times, the AI is still hired to do the work and slowly tweaks the final results.“Some people find it a bit of
=== GENERATED SEQUENCE 38 ===
TITLE= Uni revealed it's taking the wraps off its AI-powered persono at this year's EVO TEXT= Technology has made headlines this year as the virtual reality (VR) industry takes off. But the tech to interact with customers, such as NVIDIA’s GeForce graphics-rendering tech, has certainly found a niche in some devices running virtual reality apps.Now Uni is releasing the full details of its AI persono — the neural network-powered persono that’s supposed to challenge the competition at EVO 2020. Uni’s not the only brand looking to test AI out of CES, but it’s demonstrating the virtue of depth, and maybe designing something a bit bigger.The Uni product was unveiled at the recent EVO event in Las Vegas, and showed a woman who appeared to be in the crowd dressed as an Amazon Alexa virtual assistant. Uni even showed off a virtual robotic arm that can shoot guns. We’ve reached out to the brand to learn more, but are awaiting updates.“NU is aware of
=== GENERATED SEQUENCE 39 ===
TITLE= Government should make it easier to sue people who confuse AI TEXT= SUSP researchers have developed a new obstacle: Computers that confuse AI are now illegal everywhere, from schools to law enforcement. The problem has come up across tech companies in the past few years, but no longer represents a serious threat to humanity. It’s an issue with big tech’s main roadblocks: sensors that monitor our faces and bodies.But such issues are just as important as a lawsuit, because they can impact people in ways that can go unreported or are justifiably controversial. In recent years, some tech companies have used AI to stoke fears about general intelligence—enormously high levels of self-awareness and advanced planning that let an algorithm anticipate an event horizon that someone else would not anticipate. Last year, for example, Facebook announced that it was taking steps to try to sue an AI that taught itself everything from recognizing the faces of all passengers to taking the names of its over-sharing customers.This time around, the tech company tried a different approach: It was less conceptual in tone, deploying a technical solution that, without too much exposition, showed few limits. The approach was so subtle, in fact, that some experts who
=== GENERATED SEQUENCE 40 ===
TITLE= Measuring abstract knowledge without writing it TEXT= I’m 10 years old. I have a lot of expertise in computer science, and I want to use that to measure my understanding of knowledge. If I can show that I know what a book is about, and what’s in it, then it can be done. I’m hoping to use this to measure human knowledge without writing it. I’m very familiar with trying various kinds of natural language questions, but with no idea how to begin. I've been meaning to program an AI program (a bot) to do this.This program does what I think it should be able to do (a summary of the major terms it thinks are commonly understood):It only needs a basic knowledge of the word "ahhhh" to understand that it's a bird. But apparently it can also answer questions about the Bible, using only its knowledge of the scientific language to fill in the gaps.Hence it can probably do the following:Write a short story about a book that contains a number of other terms that relate to the Bible, and fill in the blanks using only their knowledge.If I have all the information in my head, I could do the following,
=== GENERATED SEQUENCE 41 ===
TITLE= ACLU Statement on Robotic Trucks TEXT= In January of 2019, California banned government agencies from the use of facial recognition technologies, effectively curbing their own role in biometric surveillance. The ACLU has consistently supported driverless vehicle bans, but it finally made its way Tuesday, when the group held a series of smaller rallies across the country and petitioned the state’s legislatures to stop their implementation.Here’s how the groups worked:Over a year ago, PAVE—a nonprofit that creates safety, policy, and awareness campaigns to increase public awareness about autonomous vehicle safety—published a report defining automated vehicle deployment, committing driverless safety, and making policy recommendations. PAVE specifically targeted safety-critical mobility and entertainment technologies, as well as data management and situational awareness, as key topics encountered when AVs are on the road. The group specifically targeted the video cameras ubiquitous in commercial and ride-hailing fleets, as well as emergency and ride-hailing drivers, and the Internet of Things (IoT) and mobile apps.The group specifically
=== GENERATED SEQUENCE 42 ===
TITLE= Dungeon crawling has officially arrived at the park! TEXT= The first ever truly stand-alone adventure game is here. And it is what we get to explore as you wander the halls of this mysterious and terrifying hotel. If you've ever played a video game, you may remember its name as the name of one of the greatest video game demonstrations of the age: Montezuma’s Revenge. In 1997, at a party at a hotel, you explore a modern-day version of the classic game in a small, nondescript house as you explore to find all keys, furniture and other goodies hidden throughout the house. If you remember, this version of the game first appeared in 1997 as a text adventure, a kind of puzzle-platformer with the fewer puzzle pieces and the easier way to complete them. Montezuma’s Revenge is perhaps the most famous video game in the world.Released for PC in 1995 and played by 4K graphics and an average-beefy 60 frames per second, it spawned a wave of startups and publishers eager to push the genre forward. Now, however, a newer co-op phenomenon is creating the first real open-ended video game where the desire to explore and solve the game runs
=== GENERATED SEQUENCE 43 ===
TITLE= Can artificial intelligence fight elderly gamers? Not unless you have Alzheimer's TEXT= Can artificial intelligence fight elderly gamers? Not unless you have Alzheimer's.While there are already some ways that artificial intelligence could work out which players are at greater risk of developing Alzheimer's simply by watching YouTube games, that hasn't been possible until now. Artificial intelligence (AI) just can't, which is why it's been shown that it's possible for the software to combat the cognitive impairment it was created to combat.In research published in the prestigious journal Science Medicine, researchers show that training an AI system to play a game, such as Breakout, can actually teach it new tricks such as turning on certain graphics or phrases to make it claim that it's seen a certain movie screen.The system has already learned to recognize a phrase such as 'playing golf' in videos it's seen before, and responded with a growing number of eye exercises to demonstrate its abilities.While it's still very early days, the researchers posit that, using demonstrations to prove that AI can help with certain cognitive impairment symptoms might be a promising alternative to treatments which are currently limited by the number of symptoms it shows.If nothing else, the research proves that AI needs to be trained to play a
=== GENERATED SEQUENCE 44 ===
TITLE= Quantifying Independently Reproducible Machine Learning TEXT= Peer review has come a long way in recent years. With social media and the internet, social media has become a way of communicating and seeing each new Facebook news article as part of the person’s life — all in the knowledge that the other person (or, more likely, the author) was the author of the piece. It’s an easy way to establish credibility, generate clickbait articles, and get a new person’s online approval. But for many people, reproducibility means replicating results simply not having existed for one moment.A new paper from researchers at Stanford and Brown University offers a new way to do this. It proposes to take this sort of communication tool and make it genuinely reproducible by hiding it behind a data set of simply labeled data sets.The idea is not to create algorithms that tell you everything you need to know, but to get “something precise” about each labeled data set by collecting it and labeling it itself so the other person could perform the task. The data set would have labels such as “news articles” or “comments” — each of these could be one of many hundreds of one
=== GENERATED SEQUENCE 45 ===
TITLE= Robotics and the future of work TEXT= If you can see a future in which your job is to help make new progress in robotics, you will be hard-pressed to say not human.The future of work may not look a lot like the past — a future in which people rely on platforms like Amazon Mechanical Turk to meet human-like demands. But that future is likely to look a lot different from the dystopian one we are in now, because many roles will be replaced, many roles will be retrained, and a whole generation of workers will need to find new and better ways to join and flourish.For many of them, retraining comes with a cost — a job search online has to involve tasks like finding information about your country of residency and bringing a laptop with you when you return home. And for others, working with robots takes only a few hours of time, which can be expensive, requiring compensation that is separate to the job.“Online jobs are going to be fully automated.”Arthur Julian-BorchakovskyIt is hard for many people to find jobs in the near future when they can no longer think, work and sleep. To help fill these jobs, many try to find new and better jobs.
=== GENERATED SEQUENCE 46 ===
TITLE= The man who got deepfaked to death wanted to prove that deepfakes can be faked TEXT= US researchers at the University of Texas at Austin have developed a deepfake-generating software that can convincingly fool facial-recognition systems with convincing images.Since 2013, a team of more than 70 experts has used deepfake-generating software to create fake videos of celebrities, politicians, and corporate corporate representatives speaking in unison to alter video—including key moments in the presidential election.The system first creates a video of a speaker whose face and voice is superimposed on another person's. The system then creates a fake video with the words “Deepfakes” in the center. In most cases, the system replaces the original speaker with someone else.The system has been used to achieve good results in a limited form—including faked suicide vids and headshots. The researchers demonstrated the system to researchers and presented it with more realistic video of leading members of Congress and corporate entities.“Our work demonstrates the value of synthesizing human voices with deepfakes,” says PhD student Ryan Sinnet, who is also the paper’s senior research scientist.In experiments, the team fed the deepfake-gener
=== GENERATED SEQUENCE 47 ===
TITLE= VeriPol, a Chinese artificial intelligence company promising to predict the rest of the human genome TEXT= VeriPol, a Chinese AI company promising to predict the rest of the human genome at a cost of up to $2 billion over the next five years, has started selling its AI-powered genetic code. The software works by analyzing thousands of human genome sequences and compares these to reference sequences from around the globe to fill in the blanks. The software can boost the accuracy of in-depth genetic study, with provable reads accounting for up to 96% of the genetic variation in “geographic diversity” (see “Here’s why so few non-geographic people have 'grit'), according to Wired.The software is the latest project of a nascent industry of artificial intelligence (AI) tools. Provoking headlines and claims about superhuman computation power and massive benefits in terms of the cost savings on routine healthcare procedures, for example, VeriPol CEO Takeshi Kitano described such tools as “the next generation of diagnostic tools�
=== GENERATED SEQUENCE 48 ===
TITLE= The Evolution of AlphaStar TEXT= DeepMind, the Alphabet-backed artificial intelligence research company, has produced a machine learning agent that can distinguish a human from a bot in a StarCraft II match.Supercomputer StarCraft II champion FanTaSy | Getty ImagesOpenAI Five | Getty ImagesOpenAI, the AI research team, which was created last January to compete in the Atari video game StarCraft 2, has made a StarCraft II bot capable of beating a professional player in a virtual two-player match. But its achievements aren’t limited to locating a human at the start of a game, in this case. In a two-player match of the popular video game StarCraft, only one of four bots can reach full strength. To qualify for promotion to the e-sports tier, teams must first beat a professional player in StarCraft II. The other three rounds of competition are random, requiring either a player to choose one of eight OpenAI Five bots for each game, or a random selection of three bots for each game, to qualify for promotion. If a player loses his spot, OpenAI Five unpacks and trains to win, just like human players do, using only five bots instead of the many learned from during rounds of competition. Since most bots
=== GENERATED SEQUENCE 49 ===
TITLE= Privacy Request Forms TEXT= Privacy Request FormsFor the first time, the Illinois Secretary of State has asked the Commerce Department to create a task force to study how personal data is collected and used by the state.The State Department’s AI Impact Challenge is open to the public and invited responses, and the commission is calling to discuss the challenges ahead. The report is slated to be published on Sept. 17. More than 15,000 respondents to its survey will have 60 days to comment on the full details of the survey and request public comment on a final report.The two-year-old I.R.U. study began when Liz O’Sullivan arrived at Cook County Community Corrections in 2008. She was assigned to the aging population of O’Hare Central Regional Corrections in Chilesville, where she was the chief and chief justice. Prior to that, O’Hare used a combined 17,000-person workforce of "enhanced supervision and corrections."The feedback from the residents of Cook County prompted the commissioners to create a task force to study privacy and artificial intelligence impacts of ICE detention and adoption and decide how Illinois should regulate AI. The group’s report is due to be released next year.“
=== GENERATED SEQUENCE 50 ===
TITLE= DeepMind’s Dreamer AI learns from the past to predict the future TEXT= In AI, there’s a humongous body of knowledge. For example, 85% of scientists working on deep learning predict failure in about 90% of the cases where a model delivers a predicted answer. Dreamer, from Google’s AI subsidiary, develops predictive analytics technology that anticipates catastrophic failures in about 90% of the cases.Building on Google’s work in the academic community, Google Research (FAIR) today announced an AI system called Dreamer that provides businesses and governments with a novel way to predict which products will fail in a given circumstance. The system uses machine learning and deep learning to extract key information from past predictions, leverage correlations across multiple sources, and assess minute-to-minute context to predict failure in increasingly complex models.Using a neural network, Dreamer anticipates failure in roughly 90% of the cases. Leveraging natural language processing and reinforcement learning (RL), the system automatically learns the business-critical components of a model