Discover more from Last Week in AI
Last Week in AI #148: AI art generator app, new wave of AI language startups, facial recognition persists despite bans
New AI app generates art based on user-given styles and text prompts, new AI startups leverage open-source NLP tools, facial recognition is becoming more popular despite bans, and more!
Canadian startup Wombo.ai has recently released the app Dream for iOS and Android, which lets users create original “artworks” by just putting in a text prompt and choosing a visual style. The app is powered by AI techniques that have emerged over the course of 2021, with OpenAI DALL-E showing the potential of such an approach in early January:
As the blog post Alien Dreams: An Emerging Art Scene covers, while OpenAI did not release DALL-E for public use, they did release CLIP, a model trained to determine how well some text matches with a given image. Hackers soon found ways to combine this model with image-generating GANs to create images from text, and posting photos made with tools based on these techniques soon became popular among AI researchers and developers on Twitter.
Dream is not the first such service (NightCafe is an earlier product), but it recently became viral due to users posting their creations on TikTok, Instagrams, and other platforms.
Our take: we’ve been following this trend for a while, and it’s super fun to see people outside of AI discover what these techniques can do! It’s very likely we’ll see many more tools like this get released in 2022, and I very much look forward to it. Just for fun, here is some sample art I generated myself earlier this year:
The release of GPT-3 API was an important step in the democratization of NLP technologies. By taking the massive pretrained model and adapting it to a specific task, developers could leverage powerful NLP technology without immense computational power or deep expertise. While OpenAI's API is pay-to-play past a certain point, groups like EleutherAI have sought to make GPT-style models open-source. Of course, making such a model available requires a provider. HuggingFace is likely the most famous, but a number of other startups have arisen to serve models like Eleuther's via API: these include NLP Cloud, Neuro, and CoreWeave. These startups claim they are making NLP technology more accessible than better-funded rivals and may just drive the next wave of NLP adoption.
Our take: I think the democratization of technologies that previously required previous expertise is always an interesting development. As the VentureBeat article points out, using these models isn't always an unqualified good due to issues like bias. But I think more open use can help us more quickly understand the issues with models like GPT-3 while finding different use-cases.
While recent public policy has banned police use of facial recognition in places like San Francisco and Washington State, the use of facial recognition software by other entities has only become more prevalent, especially in travel. For example, the number of airports using facial recognition for identity checking skyrocketed from 14 in 2020 to 182 in 2021. While a patchwork of regulations has sprung up in local governments over the past couple of years, facial recognition remains largely unregulated at the federal level. Opponents of the technology point to its privacy and bias concerns, while proponents claim that the focus on the harm of police use of facial recognition has given the field a bad name.
Our take: There are legitimate use-cases but also legitimate concerns with facial recognition. For better or worse, as the technology matures and becomes easier to use, it is likely that facial recognition will only become more and more common in everyday life. Timely federal-level regulations are needed to protect consumers with broad applications of such a technology that is relatively easy to abuse.
DARPA invests in AI that can translate instruction manuals into augmented reality - "The Defense Advanced Research Projects Agency has issued a $5.8 million contract to a team building an artificial intelligence system able to scan instruction manuals and convert that data into instructions for augmented reality systems."
2021 was the year of monster AI models - "When OpenAI released GPT-3, in June 2020, the neural network’s apparent grasp of language was uncanny. It could generate convincing sentences, converse with humans, and even autocomplete code. GPT-3 was also monstrous in scale—larger than any other neural network ever built."
Is artificial intelligence about to transform the mammogram? - "An MIT researcher has devised a radical technique: an AI that can predict breast cancer. But it’s a long climb from the lab to the real world."
Bigger’s Not Always Better: DeepMind’s New Language AI Is Small But Mighty - "Bigger is better—or at least that’s been the attitude of those designing AI language models in recent years. But now DeepMind is questioning this rationale, and says giving an AI a memory can help it compete with models 25 times it ssize."
Memory Chips That Compute Will Accelerate AI - "John von Neumann’s original computer architecture, where logic and memory are separate domains, has had a good run. But some companies are betting that it’s time for a change."
OpenAI Releases GLIDE: A Scaled-Down Text-to-Image Model That Rivals DALL-E Performance - "Text-to-image generation has been one of the most active and exciting AI fields of 2021. In January, OpenAI introduced DALL-E, a 12-billion parameter version of the company’s GPT-3 transformer language model designed to generate photorealistic images using text captions as prompts."
The Future of Robot Nannies - "Childcare is the most intimate of activities. Evolution has generated drives so powerful that we will risk our lives to protect not only our own children, but quite often any child, and even the young of other species."
AI Dungeon’s creators are launching an experimental AI-powered game platform - "Latitude, the startup behind text game AI Dungeon, is expanding into a new artificial intelligence-powered game platform called Voyage. The company announced the closed beta on Friday, opening a waitlist for current AI Dungeon users."
New AI and free press standards fortify the fight against autocracy - "UNESCO announced a remarkable consensus agreement among 193 member states creating the first-ever global standard on the ethics of artificial intelligence (AI). "
Characters for good, created by artificial intelligence - "As it becomes easier to create hyper-realistic digital characters using artificial intelligence, much of the conversation around these tools has centered on misleading and potentially dangerous deepfake content."
How AI and robotics are reconstructing a 2,000-year-old fresco in Pompeii - "Computer scientists and archeologists are working together to solve this ancient puzzle. 4 minute ReadWhen we sit down to solve a jigsaw puzzle, there’s always one thing we take for granted: the picture on the box."
From Machine Learning to Sensor Systems, Food Tech Is Leading the Fight Against Food Waste - "Food waste is endemic across the supply chain, with staggering impacts for our climate."
This Automatic Skin Retouching Software Could Be a Game-Changer Thanks to AI - "When Unmesh Dinda of PiXimperfect says that this is hands down the best automated skin retouching software ever created, you might want to pay attention."
The US Military’s Future: Waves of Killer Drone Swarms - "300-Plus Swarm Drones Just Flew In This DARPA Test: One drone armed with a missile or a bomb can take out a tank or a building. Make it 10 drones, and you can create havoc on a battlefield or a city street. But what happens when 300 drones attack?"
Elon Musk says Tesla doesn't get 'rewarded' for lives saved by its Autopilot technology, but instead gets 'blamed' for the individuals it doesn't - "Tesla may be paving the way for autonomous electric vehicles in the auto industry, but that still comes with criticism over the technology's failures."
Toronto suspends self-driving bus pilot after disastrous Whitby crash - "The City of Toronto has suspended its trial of a self-driving bus after a similar one crashed into a tree in Whitby, critically injuring the onboard attendant."
Chinese scientists develop AI ‘prosecutor’ that can press its own charges - "The AI “prosecutor” can file a charge with more than 97 per cent accuracy based on a verbal description of the case, according to the researchers."
Worried about super-intelligent machines? They are already here - "Forget about the danger of robots creating a sci-fi-style dystopia. The modern corporation is already doing all of that"
Kids Grok AI, But Not Its Pitfalls - "They need a crash course in algorithmic bias"
A Move for 'Algorithmic Reparation' Calls for Racial Justice in AI - "Forms of automation such as artificial intelligence increasingly inform decisions about who gets hired, is arrested, or receives health care."
Artificial Intelligence Was the Most Discussed Technology of 2021 - "Using data from the analytics tool Buzzsumo, Walkme.com has compiled a list of the most discussed technologies online in 2021, with artificial intelligence topping the list with 175,837 online articles."
White House AI Initiative Launches Public Research Support Tools - "A new section of resources intended for artificial intelligence researchers was launched last Friday by the National Artificial Intelligence Initiative Office, with a goal of providing easy access to data sets and testbed environments for AI application training."
Senate Approves Boost in Training for Federal Employees on AI - "As one of its last actions before recessing through the holidays, the Senate approved a bill (S-2551) to boost training for federal employees who acquire and manage artificial intelligence technology on its capabilities and risks."
Campuses, Law Enforcement Reckon with Facial Recognition Bans - "Implementing facial recognition is becoming more cumbersome as an increasing number of cities and states restrict the use of this technology. Several states have enacted some conditional use restrictions of facial recognition."
Copyright © 2021 Skynet Today, All rights reserved.