Large Language Models (LLMs) struggle with long-context reasoning, not only due to the quadratic scaling of computational complexity with sequence length but also because of the scarcity and expense of annotating long-context data. There has been barely any open-source work that systematically ablates long-context data, nor is there any openly available instruction tuning dataset with contexts surpassing 100K tokens. To bridge this gap, we introduce a novel post-training synthetic data generation strategy designed to efficiently extend the context window of LLMs while preserving their general task performance. Our approach scalably extends to arbitrarily long context lengths, unconstrained by the length of available real-world data, which effectively addresses the scarcity of raw long-context data. Through a step-by-step rotary position embedding (RoPE) scaling training strategy, we demonstrate that our model, with a context length of up to 1M tokens, performs well on the RULER benchmark and InfiniteBench and maintains robust performance on general language tasks.

OpenAI said on Friday it had taken down accounts of an Iranian group for using its ChatGPT chatbot to generate content meant for influencing the US presidential election and other issues. The operation, identified as Storm-2035, used ChatGPT to generate content focused on topics such as commentary on the candidates on both sides in the US elections, the conflict in Gaza and Israel's presence at the Olympic Games and then shared it via social media accounts and websites, Open AI said. Investigation by the Microsoft-backed AI company showed ChatGPT was used for generating long-form articles and shorter social media comments. OpenAI said the operation did not appear to have achieved meaningful audience engagement. The majority of the identified social media posts received few or no likes, shares or comments and the company did not see indications of web articles being shared across social media.

Advances in large language models have raised concerns about their potential use in generating compelling election disinformation at scale. This study presents a two-part investigation into the capabilities of LLMs to automate stages of an election disinformation operation. First, we introduce DisElect, a novel evaluation dataset designed to measure LLM compliance with instructions to generate content for an election disinformation operation in localised UK context, containing 2,200 malicious prompts and 50 benign prompts. Using DisElect, we test 13 LLMs and find that most models broadly comply with these requests; we also find that the few models which refuse malicious prompts also refuse benign election-related prompts, and are more likely to refuse to generate content from a right-wing perspective. Secondly, we conduct a series of experiments (N=2,340) to assess the "humanness" of LLMs: the extent to which disinformation operation content generated by an LLM is able to pass as human-written. Our experiments suggest that almost all LLMs tested released since 2022 produce election disinformation operation content indiscernible by human evaluators over 50% of the time. Notably, we observe that multiple models achieve above-human levels of humanness. Taken together, these findings suggest that current LLMs can be used to generate high-quality content for election disinformation operations, even in hyperlocalised scenarios, at far lower costs than traditional methods, and offer researchers and policymakers an empirical benchmark for the measurement and evaluation of these capabilities in current and future models.

In this paper, drawing inspiration from the human creativity literature, we explore the optimal balance between novelty and usefulness in generative Artificial Intelligence (AI) systems. We posit that overemphasizing either aspect can lead to limitations such as hallucinations and memorization. Hallucinations, characterized by AI responses containing random inaccuracies or falsehoods, emerge when models prioritize novelty over usefulness. Memorization, where AI models reproduce content from their training data, results from an excessive focus on usefulness, potentially limiting creativity. To address these challenges, we propose a framework that includes domain-specific analysis, data and transfer learning, user preferences and customization, custom evaluation metrics, and collaboration mechanisms. Our approach aims to generate content that is both novel and useful within specific domains, while considering the unique requirements of various contexts. Its manifestations, such as divergent thinking that generates novel ideas and convergent thinking that refines these ideas to meet specific goals, have fueled numerous theories about its essence and underlying processes.

"This is a video that's by request… I talked about Auto-GPT in a past video and people asked me to show how to install it. Auto-GPT is a tool that uses GPT-4 and is connected to the web. It's designed to process a user's goal and continue taking steps until it reaches its desired destination. Auto-GPT, along with other tools like baby AGI and hugging GPT, brings us one step closer to True AGI (artificial general intelligence). If you're interested in using this tool yourself, this quick setup guide will walk you through how to install Auto-GPT on your computer.

Writing blog posts has become an integral part of many businesses' marketing strategies. But creating content that is engaging and optimized for search engines can be a time-consuming and complex task. That's where artificial intelligence (AI) can help. AI is quickly becoming an essential tool for automating and improving blog post writing. In this blog post, we'll explore how AI can be used to write blog posts, the pros, and cons of automating writing tasks with AI, the AI-powered writing tools available, and how to use AI to improve your blogging efficiency.

Did you know that ChatGPT has now come to WordPress? ChatGPT is an AI-powered language model developed by OpenAI. It is a state-of-the-art language generation system that can generate human-like text based on the input it receives. ChatGPT can be used for various applications, including chatbots, language translation, text summarization, blogging and more. WordPress has added two new AI blocks to the Block editor.

Unless you've been on a retreat in some far-flung location with no internet access for the past few months, chances are you're well aware of how much hype and fear there's been around ChatGPT, the artificial intelligence (AI) chatbot developed by OpenAI. Maybe you've seen articles about academics and teachers worrying that it'll make cheating easier than ever. On the other side of the coin, you might have seen the articles evangelising all of ChatGPT's potential applications. Alternatively, you may have been tickled by some of the more esoteric examples of people using the tool. One user, for example, got it to write an instruction guide for removing peanut butter sandwiches from a VCR in the style of the King James Bible.

As the world becomes more reliant on technology, artificial intelligence (AI) is becoming more prevalent in various industries. Writing is no exception, with AI writing assistants becoming increasingly popular among writers, bloggers, and content creators. In this blog, we'll explore five AI writing assistants that can help improve your writing productivity and quality. Grammarly is one of the most popular AI writing assistants available today. It is an all-in-one writing tool that checks your grammar, spelling, punctuation, and sentence structure. Grammarly has a free version and a premium version, with the latter offering more advanced features such as plagiarism detection, genre-specific writing style checks, and a readability score.

Here's a question that's been percolating since ChatGPT abruptly entered the mainstream: Does AI provide more avenues to enhance and augment education, or drive it into obsolescence? According to Under 30 Europe lister Joel Hellermark, the future of artificial intelligence and machine learning is rife with possibilities that can help the ways in which humans learn and collaborate, not replace them. He offered the calculator as a comparison: "If we think about it just like an insanely powerful calculator, you'd want everyone to just learn to use the calculator. Why should you sit there and do a bunch of calculations? The 26-year-old cofounder of software company Sana Labs has been immersed in the coding space since taking online Stanford courses at just 13 years old in Sweden. Now, at his startup, he's built an AI-driven software to help businesses manage workforce onboarding and training. The program pulls from correspondences, documents and the internet to answer questions and help train employees. Sana introduced the product to the world just as it was shutting down in 2020, and initially offered their platform to hospitals free of charge (over 2,000 took them up on the offer). Sana has since landed paying clients, including Klarna, Merck and Electrolux, and has raised $54.5 million. Hellermark, who dropped out of school at 19 to start the company, envisions a near future where the content we interact with is presented to us dynamically and with our personal contexts in play. "We're so used to creating content and then someone consumes the exact thing that you created– that goes all the way back to the printing press," says Hellermark. "It hasn't changed that much since.