Last year, Rachel Antell, an archival producer for documentary films, started noticing AI-generated images mixed in with authentic photos. There are always holes or limitations in an archive; in one case, film-makers got around a shortage of images for a barely photographed 19th-century woman by using AI to generate what looked like old photos. Which brought up the question: should they? And if they did, what sort of transparency is required? The capability and availability of generative AI – the type that can produce text, images and video – have changed so rapidly, and the conversations around it have been so fraught, that film-makers' ability to use it far outpaces any consensus on how.

Y Combinator is famed for its Demo Days, where portfolio companies pitch their apps and wares in hopes of growing from a fledgling company into the next AirBnB. But on Thursday, the startup incubator hosted a mélange of founders, venture capitalists, and US policy makers in its airy industrial space in San Francisco to tackle a defining topic for so many startups today: AI as the latest frontier in the battle between Big Tech and the little guys. For many early-stage tech entrepreneurs, questions around AI can carry existential weight. Ever since ChatGPT was unleashed in late 2022, OpenAI's technology, along with fast follows from Google's and Microsoft's AI teams, has dominated the conversation around this new era of artificial intelligence. But the increasing availability--and potency--of open source AI models has the potential to upend those dynamics.

Apple has joined several other tech companies in agreeing to abide by voluntary AI safeguards laid out by the Biden administration. Those who make the pledge have committed to abide by eight guidelines related to safety, security and social responsibility, including flagging societal risks such as biases; testing for vulnerabilities, watermarking AI-generated images and audio; and sharing trust and safety details with the government and other companies. Amazon, Google, Microsoft and OpenAI were among the initial adoptees of the pact, which the White House announced last July. The voluntary agreement, which is not enforceable, will expire after Congress passes laws to regulate AI. Since the guidelines were announced, Apple unveiled a suite of AI-powered features under the umbrella name of Apple Intelligence.

OpenAI announced a new AI-powered search engine prototype called SearchGPT. It's described SearchGPT as "a temporary prototype of new AI search features that give you fast and timely answers with clear and relevant sources." The company plans to test out the product with 10,000 initial users, then roll it into ChatGPT after gathering feedback. It's a spicy time to launch AI-powered search engines. Last month, Perplexity faced criticism for summarizing stories from Forbes and Wired without adequate attribution or backlinks to the publications.

Google executives may be fearing the worst once again as Microsoft-backed rival OpenAI launches a new AI-powered search tool. 'SearchGPT', which is being trialed as a prototype before a wider rollout, scours the web for live news and information just like Google Search. OpenAI says the new product is particularly useful for queries about current events, recent developments, or specific information that ChatGPT might not know. Social media users have noted the parallels with the world's biggest search engine, with one saying'Google Search is definitely in trouble'. Another said: 'Anyone who has been paying attention knows there will be a new king of search within 10 years.

You might be familiar with such AI tools--they're the ones that can create text, images, or even films from various prompts. Do our genomes really work in the same way? When I was at school, I was taught that the genome is essentially a code for an organism. It contains the instructions needed to make the various proteins we need to build our cells and tissues and keep them working. It made sense to me to think of the human genome as being something like a program for a human being.

Amritha R Warrier & AI4Media / Better Images of AI / error cannot generate / Licenced by CC-BY 4.0 The rapid rise of generative artificial intelligence like OpenAI's GPT-4 has brought remarkable advancements, but it also presents significant risks. One of the most pressing issues is model collapse, a phenomenon where AI models trained on largely AI-generated content tend to degrade over time. This degradation occurs as AI models lose information about their true underlying data distribution, resulting in increasingly similar and less diverse outputs full of biases and errors. As the internet becomes flooded with real-time AI-generated content, the scarcity of new, human-generated or natural data further exacerbates this problem. Without a steady influx of diverse, high-quality data, AI systems risk becoming less accurate and reliable.

Retrieval augmented generation (RAG) enhances the accuracy and reliability of generative AI models by sourcing factual information from external databases, which is extensively employed in document-grounded question-answering (QA) tasks. Off-the-shelf RAG flows are well pretrained on general-purpose documents, yet they encounter significant challenges when being applied to knowledge-intensive vertical domains, such as electronic design automation (EDA). This paper addresses such issue by proposing a customized RAG framework along with three domain-specific techniques for EDA tool documentation QA, including a contrastive learning scheme for text embedding model fine-tuning, a reranker distilled from proprietary LLM, and a generative LLM fine-tuned with high-quality domain corpus. Furthermore, we have developed and released a documentation QA evaluation benchmark, ORD-QA, for OpenROAD, an advanced RTL-to-GDSII design platform. Experimental results demonstrate that our proposed RAG flow and techniques have achieved superior performance on ORD-QA as well as on a commercial tool, compared with state-of-the-arts. The ORD-QA benchmark and the training dataset for our customized RAG flow are open-source at https://github.com/lesliepy99/RAG-EDA.

In modern healthcare, addressing the complexities of accurate disease prediction and personalized recommendations is both crucial and challenging. This research introduces MLtoGAI, which integrates Semantic Web technology with Machine Learning (ML) to enhance disease prediction and offer user-friendly explanations through ChatGPT. The system comprises three key components: a reusable disease ontology that incorporates detailed knowledge about various diseases, a diagnostic classification model that uses patient symptoms to detect specific diseases accurately, and the integration of Semantic Web Rule Language (SWRL) with ontology and ChatGPT to generate clear, personalized health advice. This approach significantly improves prediction accuracy and ensures results that are easy to understand, addressing the complexity of diseases and diverse symptoms. The MLtoGAI system demonstrates substantial advancements in accuracy and user satisfaction, contributing to developing more intelligent and accessible healthcare solutions. This innovative approach combines the strengths of ML algorithms with the ability to provide transparent, human-understandable explanations through ChatGPT, achieving significant improvements in prediction accuracy and user comprehension. By leveraging semantic technology and explainable AI, the system enhances the accuracy of disease prediction and ensures that the recommendations are relevant and easily understood by individual patients. Our research highlights the potential of integrating advanced technologies to overcome existing challenges in medical diagnostics, paving the way for future developments in intelligent healthcare systems. Additionally, the system is validated using 200 synthetic patient data records, ensuring robust performance and reliability.

Generative AI, especially via Large Language Models (LLMs), has transformed content creation across text, images, and music, showcasing capabilities in following instructions through prompting, largely facilitated by instruction tuning. Instruction tuning is a supervised fine-tuning method where LLMs are trained on datasets formatted with specific tasks and corresponding instructions. This method systematically enhances the model's ability to comprehend and execute the provided directives. Despite these advancements, LLMs still face challenges in consistently interpreting complex, multi-step instructions and generalizing them to novel tasks, which are essential for broader applicability in real-world scenarios. This article explores why neurosymbolic AI offers a better path to enhance the instructability of LLMs. We explore the use a symbolic task planner to decompose high-level instructions into structured tasks, a neural semantic parser to ground these tasks into executable actions, and a neuro-symbolic executor to implement these actions while dynamically maintaining an explicit representation of state. We also seek to show that neurosymbolic approach enhances the reliability and context-awareness of task execution, enabling LLMs to dynamically interpret and respond to a wider range of instructional contexts with greater precision and flexibility.