We introduce a language generation task grounded in a popular video game environment. KNUDGE (KNowledge Constrained User-NPC Dialogue GEneration) requires models to produce trees of dialogue between video game characters that accurately reflect quest and entity specifications stated in natural language. KNUDGE is constructed from side quest dialogues drawn directly from game data of Obsidian Entertainment's The Outer Worlds, leading to real-world complexities in generation: (1) dialogues are branching trees as opposed to linear chains of utterances; (2) utterances must remain faithful to the game lore -- character personas, backstories, and entity relationships; and (3) a dialogue must accurately reveal new quest details to the human player. We report results for a set of neural generation models using supervised and in-context learning techniques; we find competent performance but room for future work addressing the challenges of creating realistic, game-quality dialogues.

The race to roll out artificial intelligence is happening as quickly as the race to contain it – as two key moments this week demonstrate. On 10 May, Google announced plans to deploy new large language models, which use machine learning techniques to generate text, across its existing products. "We are reimagining all of our core products, including search," said Sundar Pichai, the CEO of Google's parent company Alphabet, at a press conference. The move is widely seen as a response to Microsoft adding similar functionality to its search engine, Bing. A day later, politicians in the European Union agreed on new rules dictating how and when AI can be used.

To avoid any terminological doubt, when we write "models of the language", we refer to any models that explain some linguistic phenomena, while "language models" refer to probabilistic language models as defined in Subsection 2.3 Probabilistic Language Models. While not long ago probabilistic language models were just models that assign probabilities to sequences of words [4], now they are the cornerstone of any task in computational linguistics through few-shot learning [6], prompt engineering [38] or fine-tuning [13]. On the other hand, current language models fail to catch long-range dependencies in the text consistently. For example, text generation with maximum likelihood target leads to rapid text degeneration, and consistent text generation requires probabilistic sampling and other tricks [22]. Large language models such as GPT-3 [6] push the boundary of "short text" rather far (specifically, to 2048 tokens), but do not remove the problem. Our contributions in this work are the following: We explain how the laws of autocorrelations decay in texts are related to applicability of language models to long texts; We pioneer the use of pretrained word vectors for autocorrelation computations that allows us to study a widest range of autocorrelation distances; We show that the autocorrelations in literary texts decay according to power laws for all these distances; We show that distributional semantics typically provides coherent autocorrelations decay exponents for texts translated to multiple languages, unlike earlier flawed approaches; We show that the behavior of autocorrelations decay in generated texts is quantitatively and often qualitatively different from the literary texts.

Private and public sector structures and norms refine how emerging technology is used in practice. In healthcare, despite a proliferation of AI adoption, the organizational governance surrounding its use and integration is often poorly understood. What the Health AI Partnership (HAIP) aims to do in this research is to better define the requirements for adequate organizational governance of AI systems in healthcare settings and support health system leaders to make more informed decisions around AI adoption. To work towards this understanding, we first identify how the standards for the AI adoption in healthcare may be designed to be used easily and efficiently. Then, we map out the precise decision points involved in the practical institutional adoption of AI technology within specific health systems. Practically, we achieve this through a multi-organizational collaboration with leaders from major health systems across the United States and key informants from related fields. Working with the consultancy IDEO [dot] org, we were able to conduct usability-testing sessions with healthcare and AI ethics professionals. Usability analysis revealed a prototype structured around mock key decision points that align with how organizational leaders approach technology adoption. Concurrently, we conducted semi-structured interviews with 89 professionals in healthcare and other relevant fields. Using a modified grounded theory approach, we were able to identify 8 key decision points and comprehensive procedures throughout the AI adoption lifecycle. This is one of the most detailed qualitative analyses to date of the current governance structures and processes involved in AI adoption by health systems in the United States. We hope these findings can inform future efforts to build capabilities to promote the safe, effective, and responsible adoption of emerging technologies in healthcare.

This story is adapted from Death of an Author, a murder-mystery novella written by Aidan Marchine, a collaboration between author Stephen Marche and three artificial intelligence tools: ChatGPT, Sudowrite, and Cohere. Gus Dupin, walking along the stillness of Stony Lake in the gathering night, recognized the sleek motorboat approaching his dock. A girl in a bright yellow sundress jumped off and sprinted to his mailbox, dropping in an envelope before running back. As she set off into the lake, she yelled "an honest-to-God letter" over her shoulder. Gus Dupin was not accustomed to receiving letters or messages of any kind.

As voice-based Conversational Assistants (CAs), including Alexa, Siri, Google Home, have become commonly embedded in households, many children now routinely interact with Artificial Intelligence (AI) systems. It is important to research children's experiences with consumer devices which use AI techniques because these shape their understanding of AI and its capabilities. We conducted a mixed-methods study (questionnaires and interviews) with primary-school children aged 6-11 in Scotland to establish children's understanding of how voice-based CAs work, how they perceive their cognitive abilities, agency and other human-like qualities, their awareness and trust of privacy aspects when using CAs and what they perceive as appropriate verbal interactions with CAs. Most children overestimated the CAs' intelligence and were uncertain about the systems' feelings or agency. They also lacked accurate understanding of data privacy and security aspects, and believed it was wrong to be rude to conversational assistants. Exploring children's current understanding of AI-supported technology has educational implications; such findings will enable educators to develop appropriate materials to address the pressing need for AI literacy.

Diffusion models have become a new generative paradigm for text generation. Considering the discrete categorical nature of text, in this paper, we propose GlyphDiffusion, a novel diffusion approach for text generation via text-guided image generation. Our key idea is to render the target text as a glyph image containing visual language content. In this way, conditional text generation can be cast as a glyph image generation task, and it is then natural to apply continuous diffusion models to discrete texts. Specially, we utilize a cascaded architecture (ie a base and a super-resolution diffusion model) to generate high-fidelity glyph images, conditioned on the input text. Furthermore, we design a text grounding module to transform and refine the visual language content from generated glyph images into the final texts. In experiments over four conditional text generation tasks and two classes of metrics (ie quality and diversity), GlyphDiffusion can achieve comparable or even better results than several baselines, including pretrained language models. Our model also makes significant improvements compared to the recent diffusion model.

The first thing Geoffrey Hinton says when we start talking, and the last thing he repeats before I turn off my recorder, is that he left Google, his employer of the past decade, on good terms. "I have no objection to what Google has done or is doing, but obviously the media would love to spin me as'a disgruntled Google employee'. It's an important clarification to make, because it's easy to conclude the opposite. After all, when most people calmly describe their former employer as being one of a small group of companies charting a course that is alarmingly likely to wipe out humanity itself, they do so with a sense of opprobrium. But to listen to Hinton, we're about to sleepwalk towards an existential threat to civilisation without anyone involved acting maliciously at all. Known as one of three "godfathers of AI", in 2018 Hinton won the ACM Turing award – the Nobel prize of computer scientists for his work on "deep learning". A cognitive psychologist and computer scientist by training, he wasn't motivated by a desire to radically improve technology: instead, it was to understand more about ourselves. "For the last 50 years, I've been trying to make computer models that can learn stuff a bit like the way the brain learns it, in order to understand better how the brain is learning things," he tells me when we meet in his sister's house in north London, where he is staying (he usually resides in Canada). Looming slightly over me – he prefers to talk standing up, he says – the tone is uncannily reminiscent of a university tutorial, as the 75-year-old former professor explains his research history, and how it has inescapably led him to the conclusion that we may be doomed. In trying to model how the human brain works, Hinton found himself one of the leaders in the field of "neural networking", an approach to building computer systems that can learn from data and experience. Until recently, neural nets were a curiosity, requiring vast computer power to perform simple tasks worse than other approaches. But in the last decade, as the availability of processing power and vast datasets has exploded, the approach Hinton pioneered has ended up at the centre of a technological revolution. "In trying to think about how the brain could implement the algorithm behind all these models, I decided that maybe it can't – and maybe these big models are actually much better than the brain," he says. A "biological intelligence" such as ours, he says, has advantages. It runs at low power, "just 30 watts, even when you're thinking", and "every brain is a bit different". That means we learn by mimicking others. But that approach is "very inefficient" in terms of information transfer. Digital intelligences, by contrast, have an enormous advantage: it's trivial to share information between multiple copies. "You pay an enormous cost in terms of energy, but when one of them learns something, all of them know it, and you can easily store more copies.

'Gutfeld!' panelists react to Vice President Kamala Harris leading the White House's AI meetings with the CEOs of Alphabet, Anthropic, Microsoft and OpenAI. It's official, this is now the best late night show in America, because it's the only late night show in America. So today, senior intel officials testified on Capitol Hill on worldwide threats, among the topics, China, Russia, Iran, artificial intelligence, and also Geraldo removing his shirt in front of children. Yeah, AI is now in the same discussion as some of our biggest, most dangerous adversaries. So you think we'd put someone serious in charge of it, right?

Users in consumption domains, like music, are often able to more efficiently provide preferences over a set of items (e.g. a playlist or radio) than over single items (e.g. songs). Unfortunately, this is an underexplored area of research, with most existing recommendation systems limited to understanding preferences over single items. Curating an item set exponentiates the search space that recommender systems must consider (all subsets of items!): this motivates conversational approaches-where users explicitly state or refine their preferences and systems elicit preferences in natural language-as an efficient way to understand user needs. We call this task conversational item set curation and present a novel data collection methodology that efficiently collects realistic preferences about item sets in a conversational setting by observing both item-level and set-level feedback. We apply this methodology to music recommendation to build the Conversational Playlist Curation Dataset (CPCD), where we show that it leads raters to express preferences that would not be otherwise expressed. Finally, we propose a wide range of conversational retrieval models as baselines for this task and evaluate them on the dataset.