Likelihood-based, or explicit, deep generative models use neural networks to construct flexible high-dimensional densities. This formulation directly contradicts the manifold hypothesis, which states that observed data lies on a low-dimensional manifold embedded in high-dimensional ambient space. In this paper we investigate the pathologies of maximum-likelihood training in the presence of this dimensionality mismatch. We formally prove that degenerate optima are achieved wherein the manifold itself is learned but not the distribution on it, a phenomenon we call manifold overfitting. We propose a class of two-step procedures consisting of a dimensionality reduction step followed by maximum-likelihood density estimation, and prove that they recover the data-generating distribution in the nonparametric regime, thus avoiding manifold overfitting. We also show that these procedures enable density estimation on the manifolds learned by implicit models, such as generative adversarial networks, hence addressing a major shortcoming of these models. Several recently proposed methods are instances of our two-step procedures; we thus unify, extend, and theoretically justify a large class of models.

Why it matters: Minecraft may not sound like an important tool that supports advanced AI research. After all, what could possibly be so important about teaching a machine to play a sandbox game released more than a decade ago? Based on OpenAI's recent efforts, a well-trained Minecraft bot is more relevant to AI advancement than most people might realize. OpenAI has always focused on artificial intelligence (AI) and machine learning advances that benefit humanity. Recently, the company successfully trained a bot to play Minecraft using more than 70,000 hours of gameplay videos. The achievement is far more than just a bot playing a game.

I recently caught up with Ian Hogarth and Nathan Benaich, who each year produce The State of AI Report, a must-read snapshot of how commercial applications of A.I. are evolving. Benaich is the founder of Air Street Capital, a solo venture capital fund that is one of the savviest early-stage investors in A.I.-based startups I know. Hogarth is the former co-founder of concert discovery app Songkick and has since go on to become a prominent angel investor as well one of the founders behind the founder-lead European venture capital platform Plural. There's always a lot to digest in their report. But one of the key takeaways from this year's State of AI is that concerns established tech giants and their affiliated A.I. research labs would monopolize the development of A.I. have been proven, if not exactly wrong, then at least premature. While it is true that Alphabet (which has both Google Brain and Deepmind in its stable), Meta, Microsoft, and OpenAI (which is closely partnered now with Microsoft) are building large "foundational models" for natural language processing and image and video generation, they are hardly the only players in the game.

The human mind can get really complicated at times. That's when we turn to meditation, taking deep breaths to forget all the chaos. It gives us certain fulfilment--by bringing out new traits of understanding and empathy--sometimes by not doing anything at all. What if machines could also meditate or do nothing for a day or two--to learn better? As absurd as it may sound, a group of researchers have now discovered how artificial neural networks can mimic sleep patterns of the human brain, boosting their utility across a spectrum of research areas.

Samuel L Jackon has been made into R2D2 using the DALL-E 2 artificial intelligence program. By now, many of us have seen hilarious or ridiculous images gracing our social media feed courtesy of DALL-E 2, a complex ai art generator that allows for users to create expansive images using only a basic text prompt. DALL-E 2 can create brand new images from scratch or edit existing images, often leading to disturbing or creepy distortions of people or things that seem familiar but… not quite right. For instance, Samuel L Jackson as R2-D2 provides enough visual context to make anyone aware of who is being depicted at a passing glance, but still doesn't look quite like the Mace Windu actor, or even quite human at all. Images such as this, or Kanye Light Year, provide a glimpse into the uncanny valley.

The result is a breakthrough for a technique known as imitation learning, in which neural networks are trained how to perform tasks by watching humans do them. Imitation learning can be used to train AI to control robot arms, drive cars or navigate webpages. There is a vast amount of video online showing people doing different tasks. By tapping into this resource, the researchers hope to do for imitation learning what GPT-3 did for large language models. "In the last few years we've seen the rise of this GPT-3 paradigm where we see amazing capabilities come from big models trained on enormous swathes of the internet," says Bowen Baker at OpenAI, one of the team behind the new Minecraft bot.

When the concept artist and illustrator RJ Palmer first witnessed the fine-tuned photorealism of compositions produced by the AI image generator Dall-E 2, his feeling was one of unease. The tool, released by the AI research company OpenAI, showed a marked improvement on 2021's Dall-E, and was quickly followed by rivals such as Stable Diffusion and Midjourney. Type in any surreal prompt, from Kermit the frog in the style of Edvard Munch, to Gollum from The Lord of the Rings feasting on a slice of watermelon, and these tools will return a startlingly accurate depiction moments later. Cosmopolitan trumpeted the world's first AI-generated magazine cover, and technology investors fell over themselves to wave in the new era of "generative AI". The image-generation capabilities have already spread to video, with the release of Google's Imagen Video and Meta's Make-A-Video.

Changes to Stable Diffusion are notable, as the software is hugely influential and helps set norms in the fast-moving generative AI scene. Unlike rival models like OpenAI's DALL-E, Stable Diffusion is open source. This allows the community to quickly improve on the tool and for developers to integrate it into their products free of charge. But it also means Stable Diffusion has fewer constraints in how it's used and, as a consequence, has attracted significant criticism. In particular, many artists, like Rutkowski, are annoyed that Stable Diffusion and other image generating models were trained on their artwork without their consent and can now reproduce their styles.

Google is working on a secretive project that uses machine learning to train code to write, fix, and update itself. This project is part of a broader push by Google into so-called generative artificial intelligence, which uses algorithms to create images, videos, code, and more. It could have profound implications for the company's future and developers who write code. The project, which began life inside Alphabet's X research unit and was codenamed Pitchfork, moved into Google's Labs group this summer, according to people familiar with the matter. By moving into Google, it signaled its increased importance to leaders.

Harvey, a startup building what it describes as a "copilot for lawyers," today emerged from stealth with $5 million in funding led by the OpenAI Startup Fund, the tranche through which OpenAI and its partners are investing in early-stage AI companies tackling major problems. Also participating in the round was Jeff Dean, the lead of Google AI, Google's AI research division. Harvey was founded by Winston Weinberg, a former securities and antitrust litigator at law firm O'Melveny & Myers, and Gabriel Pereyra, previously a research scientist at DeepMind, Google Brain (another of Google's AI groups) and Meta AI. Weinberg and Pereyra are roomates -- Pereyra showed Weinberg OpenAI's GPT-3 text-generating system and Weinberg realized that it could be used to improve legal workflows. "Our product provides lawyers with a natural language interface for their existing legal workflows," Pereyra told TechCrunch in an email interview. "Instead of manually editing legal documents or performing legal research, Harvey enables lawyers to describe the task they wish to accomplish in simple instructions and receive the generated result.