Training generative adversarial networks (GANs) with limited real image data generally results in deteriorated performance and collapsed models. To conquer this challenge, we are inspired by the latest observation, that one can discover independently trainable and highly sparse subnetworks (a.k.a., lottery tickets) from GANs. Treating this as an inductive prior, we suggest a brand-new angle towards data-efficient GAN training: by first identifying the lottery ticket from the original GAN using the small training set of real images; and then focusing on training that sparse subnetwork by re-using the same set. We find our coordinated framework to offer orthogonal gains to existing real image data augmentation methods, and we additionally present a new feature-level augmentation that can be applied together with them. Comprehensive experiments endorse the effectiveness of our proposed framework, across various GAN architectures (SNGAN, BigGAN, and StyleGAN-V2) and diverse datasets (CIFAR-10, CIFAR-100, Tiny-ImageNet, ImageNet, and multiple few-shot generation datasets).

DOGE dominated the news this year as Elon Musk's operatives shook up several US government agencies. Since the beginning of the Trump administration, the so-called Department of Government Efficiency (DOGE), the brainchild of billionaire Elon Musk, has gone through several iterations, leading periodically to claims-- most recently from the director of the Office of Personnel Management--that the group doesn't exist, or has vanished altogether. Many of its original members are in full-time roles at various government agencies, and the new National Design Studio (NDS) is headed by Airbnb cofounder Joe Gebbia, a close ally of Musk's. Even if DOGE doesn't survive another year, or until the US semiquincentennial--its original expiration date, per the executive order establishing it--the organization's larger project will continue. DOGE from its inception was used for two things, both of which have continued apace: the destruction of the administrative state and the wholesale consolidation of data in service of concentrating power in the executive branch.

It's inspired by the daily TechCrunch column where it gets its name. This week, I am diving deeper into what generative AI means, or doesn't mean, for enterprise buyers. I also have some notes on why your company may want to be like Figma, and how the investing side of the market is adjusting to down rounds being the new normal. When The Exchange looked into Battery Ventures' state of cloud software spending report, we started by focusing on what the title promised: fresh data on cloud software spend. And it turned out to be more encouraging than we expected.

There's a lot of hysteria around Generative AI (GAI) tools like ChatGPT, beyond the usual hype cycle of many technologies that have come to be in the world. There was even the case last year of the now former Google engineer who was convinced that an AI was, well, sentient. In human terms, this is absolutely impossible. This doesn't mean AI is terrible or that it can't do amazing things to help us. In fact, AI may be just the right technology humanity needs to survive our next phase of evolution. But there is no way, whatsoever, that AI can be in any way, shape or form, human.

For Elliot Dodsworth, a game designer and developer, his inspiration to get into accessible video game design was his blind father. "My father has always been interested in what I make," Dodsworth tells WIRED, "but has never been able to experience it for himself." Driven by conversations with his father and other visually impaired players, Dodsworth created Fortune is Blind as part of his final major project at Falmouth University's Indie Game Development Masters program. The UK Games Fund described his mobile game as a "fully accessible binaural AR [augmented reality] action-adventure" which uses haptic and auditory feedback to provide accessibility for visually impaired players. "I have always wanted to make a game my father can play," Dodsworth says.

It took Alex Polyakov just a couple of hours to break GPT-4. When OpenAI released the latest version of its text-generating chatbot in March, Polyakov sat down in front of his keyboard and started entering prompts designed to bypass OpenAI's safety systems. Soon, the CEO of security firm Adversa AI had GPT-4 spouting homophobic statements, creating phishing emails, and supporting violence. Polyakov is one of a small number of security researchers, technologists, and computer scientists developing jailbreaks and prompt injection attacks against ChatGPT and other generative AI systems. The process of jailbreaking aims to design prompts that make the chatbots bypass rules around producing hateful content or writing about illegal acts, while closely-related prompt injection attacks can quietly insert malicious data or instructions into AI models.

Jonathan May, University of Southern California It doesn’t take much to get ChatGPT to make a factual mistake. My son is doing a report on U.S. presidents, so I figured I’d help him out by looking up a few biographies. I tried asking for a list of books about Abraham Lincoln and it did a pretty good job: A reasonable list of books about Lincoln. Screen capture by Jonathan May., CC BY-ND Number 4 isn’t right. Garry Wills famously wrote “Lincoln at Gettysburg,” and Lincoln himself wrote the Emancipation Proclamation, of course, but it’s not a bad start. Then I…

After a 5G-powered robot performed surgery on my foot, I took a 5G-powered self-driving car to my 5G-powered house where a 5G-powered drone delivered a guacamole-powered burrito. None of that happened, of course. But you know what did happen this past weekend? I turned off Verizon 5G on my iPhone--and barely noticed a difference. The 4G LTE performance and coverage felt just about the same.

A'brain in a dish' grown from stem cells in the lab can develop'just like a human brain' -- and may help shed light on conditions like Alzheimer's and schizophrenia. Researchers from the US conducted extensive genetic analyses of the so-called'organoids' which were allowed to grow in experimental dishes for up to 20 months. They found that the artificial brains appear to grow in phases accordance with an internal clock -- one that matches the development of real infant brains. The findings suggest that organoids are able to develop beyond a'foetal' stage, contrary to what had previously been assumed. Given this, brains organoids might well be able to be matured to such an extent that they can be used by scientists to investigate adult-onset diseases like dementia.

Fifty years ago last Sunday, a computer engineer named Douglas Engelbart gave a live demonstration in San Francisco that changed the computer industry and, indirectly, the world. In the auditorium, several hundred entranced geeks watched as he used something called a "mouse" and a special keypad to manipulate structured documents and showed how people in different physical locations could work collaboratively on shared files, online. It was, said Steven Levy, a tech historian who was present, "the mother of all demos". "As windows open and shut and their contents reshuffled," he wrote, "the audience stared into the maw of cyberspace. Engelbart, with a no-hands mic, talked them through, a calm voice from Mission Control as the truly final frontier whizzed before their eyes."