Late last year, a Google engineer named Blake Lemoine felt certain he'd found something special. For months, Lemoine, who worked with the company's ethical AI division, had been testing Google's Language Model for Dialogue Applications, or LaMDA, from the living room of his San Francisco home. LaMDA is a hugely sophisticated chatbot, trained on trillions of words hoovered up from Wikipedia entries and internet posts and libraries' worth of books, and Lemoine's job was to ensure that the exchanges it produced weren't discriminatory or hateful. He posed questions to LaMDA about religion, ethnicity, sexual orientation and gender. The machine had some bugs -- there were a few ugly, racist impressions -- which Lemoine dutifully reported.

SHOPPING: Products featured in this article are independently selected by our shopping writers. If you make a purchase using links on this page, MailOnline will earn an affiliate commission. Amazon has you covered in their Prime Early Access Sale with fantastic offers on smart TVs, popular earbuds, laptops, iPhones, and so much more. Ending midnight tonight, electronic lovers can enjoy up to £200 off tons of devices, including eero Wi-Fi systems, Apple iPhones, eufy robot vacuums and Echo Buds - but hurry, you don't want to miss out, at these low prices, items are selling out fast. You'll need a Prime membership to get access to these impressive savings, so if you don't already have one, you can sign up for a free 30-day trial here.

The 1999 cult classic film Office Space depicts Peter's dreary life as a cubicle-dwelling software engineer. Every Friday, Peter tries to avoid his boss and the dreaded words: "I'm going to need you to go ahead and come in tomorrow." This scene is still popular on the internet nearly 25 years later because it captures troubling aspects of the employment relationship – the helplessness Peter feels, the fake sympathy his boss intones when issuing this directive, the never-ending demand for greater productivity. There is no shortage of pop culture depictions of horrible bosses. There is even a film with that title.

Mathematician Richard E. Bellman, whose Bellman equation also known as a …

Research Engineer, Machine/Deep Learning and Network Security in Electrical & Electronic Engineering, Research Related with NATIONAL UNIVERSITY OF …

Artificial intelligence will soon have a formidable presence in the arts. DALL-E 2 and Stable Diffusion have demonstrated their ability to generate complex and interesting visual images. There is already AI-generated music, and it will get better. There is even talk of dictating a story into a computer and the software generating a short digital movie. The question is whether or to what extent such art will catch on, given the diminished role of human creativity.

Does prompting a large language model (LLM) like GPT-3 with explanations improve in-context learning? We study this question on two NLP tasks that involve reasoning over text, namely question answering and natural language inference. We test the performance of four LLMs on three textual reasoning datasets using prompts that include explanations in multiple different styles. For these tasks, we find that including explanations in the prompts for OPT, GPT-3 (davinci), and InstructGPT (text-davinci-001) only yields small to moderate accuracy improvements over standard few-show learning. However, text-davinci-002 is able to benefit more substantially. We further show that explanations generated by the LLMs may not entail the models' predictions nor be factually grounded in the input, even on simple tasks with extractive explanations. However, these flawed explanations can still be useful as a way to verify LLMs' predictions post-hoc. Through analysis in our three settings, we show that explanations judged by humans to be good--logically consistent with the input and the prediction--more likely cooccur with accurate predictions. Following these observations, we train calibrators using automatically extracted scores that assess the reliability of explanations, allowing us to improve performance post-hoc across all of our datasets.

In the last two decades, the predominant rise of the social media industry has made the notion of a newsfeed an integral part of our lives. In a newsfeed, a user observes a structured sequence of content items (posts, photos etc.) particularly selected by the platform according to her/his preferences. Apart from social media, an analogous idea - potentially relabeled - also appears in different domains as, for example, "frequently bought together" in e-commerce, "shuffling similar songs" in music recommendation, or "recommended articles" in scholarly literature indexing databases. Whether it is measured in terms of click-rate or time devoted, the high-level objective of newsfeeds is fairly well-known: to maximize the user's engagement with the platform. In many applications, however, achieving this objective is not as simple as identifying the user's "favorite" content, given that her/his satisfaction can depend on the time passed since the same (or similar) content has been observed. As an example, a user's engagement can worsen if a social media feed (resp., a music recommendation platform) constantly presents content from the same source (resp., same artist). Motivated by such scenarios, researchers have recently studied online decision making problems capturing the notion of "recovering" payoffs, namely, scenarios where the payoff of an action drops (to zero) after each play and then slowly increases back to a baseline. In the context of online learning, these nonstationary models interpolate between multi-armed bandits, where the environment is assumed to be intact, and reinforcement learning, since the actions may now alter the future environment in a structured manner.

Knowledge graphs, as the cornerstone of many AI applications, usually face serious incompleteness problems. In recent years, there have been many efforts to study automatic knowledge graph completion (KGC), most of which use existing knowledge to infer new knowledge. However, in our experiments, we find that not all relations can be obtained by inference, which constrains the performance of existing models. To alleviate this problem, we propose a new model based on information retrieval and reading comprehension, namely IR4KGC. Specifically, we pre-train a knowledge-based information retrieval module that can retrieve documents related to the triples to be completed. Then, the retrieved documents are handed over to the reading comprehension module to generate the predicted answers. In experiments, we find that our model can well solve relations that cannot be inferred from existing knowledge, and achieve good results on KGC datasets.

Denoising diffusion models (DDMs) have shown promising results in 3D point cloud synthesis. To advance 3D DDMs and make them useful for digital artists, we require (i) high generation quality, (ii) flexibility for manipulation and applications such as conditional synthesis and shape interpolation, and (iii) the ability to output smooth surfaces or meshes. To this end, we introduce the hierarchical Latent Point Diffusion Model (LION) for 3D shape generation. LION is set up as a variational autoencoder (VAE) with a hierarchical latent space that combines a global shape latent representation with a point-structured latent space. For generation, we train two hierarchical DDMs in these latent spaces. The hierarchical VAE approach boosts performance compared to DDMs that operate on point clouds directly, while the point-structured latents are still ideally suited for DDM-based modeling. Experimentally, LION achieves state-of-the-art generation performance on multiple ShapeNet benchmarks. Furthermore, our VAE framework allows us to easily use LION for different relevant tasks: LION excels at multimodal shape denoising and voxel-conditioned synthesis, and it can be adapted for text- and image-driven 3D generation. We also demonstrate shape autoencoding and latent shape interpolation, and we augment LION with modern surface reconstruction techniques to generate smooth 3D meshes. We hope that LION provides a powerful tool for artists working with 3D shapes due to its high-quality generation, flexibility, and surface reconstruction. Project page and code: https://nv-tlabs.github.io/LION.