Artificial Intelligence News Live: Cornelis Networks introduces congestion-free network architecture for AI

Cornelis Networks has introduced a new 'congestion-free' network architecture designed to accelerate AI workloads, directly competing with established standards like Ethernet and InfiniBand, according to a report in IEEE Spectrum. The technology addresses the growing issue of network bottlenecks in large-scale AI training, where communication between thousands of accelerators can significantly slow down computations. By aiming to eliminate network congestion, the architecture is designed to reduce idle time for GPUs, with the company claiming it can speed up network-bound AI tasks by up to six times. Cornelis Networks, which was formed from Intel's former Omni-Path Architecture division, is positioning its solution against dominant players like NVIDIA's InfiniBand and high-speed Ethernet in the high-performance computing and data center markets.

A new paradigm in software development, referred to as "Software 3.0," views Large Language Models (LLMs) as a new type of computer, programmed through prompt engineering rather than traditional code, according to a report in ZDNet. This approach involves "vibe coding," an iterative and experimental process of refining prompts to achieve the desired behavioral output from the AI. While this method accelerates the creation of AI-powered services, the report notes that the infrastructure for the most powerful foundational models remains highly centralized with major technology companies.

Andy Konwinski, a co-founder of Databricks and Perplexity AI, has pledged $100 million of his personal funds to create a new institute for AI researchers, according to a report in TechCrunch. The new initiative is designed to provide direct funding to researchers, and it has already provided backing for a new lab led by Ion Stoica, a professor at UC Berkeley and another co-founder of Databricks.

More than one million people now have access to Amazon's new generative AI-powered Alexa+, according to a report in TechCrunch. The service is currently available for free during an Early Access period. Following its public launch, Alexa+ will be included at no additional cost for Amazon Prime subscribers, while non-Prime users will be able to subscribe to the service for $19.99 per month.

Chinese venture capital firm HongShan Capital Group (HSG) has launched Xbench, a new AI benchmark designed to address the problem of data contamination in model evaluation, according to a report in MIT Technology Review. Unlike static tests where AI models might have already seen the questions in their training data, Xbench features a constantly changing set of evaluations. This dynamic approach aims to force models to use genuine reasoning capabilities to solve novel problems, rather than simply regurgitating memorized answers. The development is intended to provide a more accurate measure of an AI's true intelligence and sidestep the issue of inflated performance scores on traditional benchmarks.

A judge has ordered OpenAI to preserve all user logs from its ChatGPT service as part of the discovery process for an ongoing class-action lawsuit, Ars Technica says in a report. The order is a litigation hold connected to the 'Hu v. OpenAI' case, which alleges the company engaged in privacy violations and unlawful data collection. The court denied a motion from users who argued the preservation order established a "mass surveillance program," clarifying the judge's role was to ensure evidence was retained for the lawsuit. OpenAI plans to challenge the ruling, which mandates the retention of extensive user data as potential evidence.

Artificial intelligence and digital solutions are being increasingly used to enhance the efficiency of global healthcare systems that are burdened by aging populations and workforce shortages, according to a report in MIT Technology Review. The technology is being applied to improve processes from diagnosis to treatment with the goal of easing pressure on strained health services. However, scaling these integrated digital health solutions faces significant hurdles, including the need for substantial investment in IT infrastructure, establishing robust data governance, and overcoming a lack of interoperability between different systems. Addressing these challenges, along with developing supportive regulatory frameworks and mitigating issues like algorithmic bias, is considered essential for the successful and equitable adoption of AI in medicine.

Leaked code indicates that xAI is developing an advanced file editor with spreadsheet support for its AI model Grok, according to a report in TechCrunch. The development signals a move by the company to compete with firms like Microsoft and Google, which offer similar AI 'copilot' features in their productivity suites. If released, this capability would expand Grok's function from a conversational AI to a practical tool for manipulating structured data within workplace applications.

Harvey AI, a company developing artificial intelligence tools for the legal sector, has seen its valuation grow to $5 billion after a new funding round co-led by Kleiner Perkins and Coatue, according to a report in TechCrunch. This development comes four months after the company raised $300 million in a Series D round that valued it at $3 billion.

Fanfiction creators are raising alarms over their work being used without permission to train generative AI models, a practice they view as a violation of community ethics against theft and monetization, The Verge says in a report. The issue was recently highlighted when a user scraped 12.6 million fanfiction stories from the online repository Archive of Our Own (AO3) and uploaded the dataset to the AI development platform Hugging Face. The action prompted an angry reaction from the fanfiction community, leading to the creation of a search tool for authors to determine if their work was included in the scraped data.

Former OpenAI Chief Scientist Ilya Sutskever has launched a new AI safety-focused company, Safe Superintelligence Inc., and is reportedly raising $2 billion in capital, PYMNTS said in a report. The new venture was co-founded by Sutskever, fellow former OpenAI researcher Jan Leike, and investor Daniel Gross. According to its founders, the firm's mission is to build a safe superintelligence as its sole focus, deliberately insulating its research from commercial pressures and product development cycles that affect other major AI labs.

India is deploying a combination of artificial intelligence and satellite-based remote sensing to map urban heat vulnerability at a highly granular, building-by-building level, as per a report in Wired. The initiative, currently active in cities such as Delhi, aims to precisely identify which structures are most susceptible to extreme temperatures. By analyzing factors like building materials, roof types, and surrounding vegetation from satellite imagery, the AI-driven system allows authorities to target relief efforts, such as the distribution of aid or the establishment of cooling centers, more effectively during heatwaves.

A new category of search engine optimization startups is emerging to help businesses ensure their content is featured and cited in AI-generated search answers, according to a report in Forbes. This development is a direct response to the increasing integration of AI summaries into platforms like Google, with its AI Overviews, and dedicated answer engines such as Perplexity. These new firms are developing a discipline known as 'Answer Engine Optimization' (AEO) or 'AI SEO', which focuses on adapting digital content strategies to be more discoverable and useful for the large language models that power these search features.

Thai authorities have issued a public warning about a sophisticated cybercrime trend involving artificial intelligence to perpetrate financial fraud, according to a report in nationthailand. The scams utilize AI voice cloning technology, also known as audio deepfakes, to realistically impersonate family members or associates in fabricated emergencies, thereby tricking victims into making urgent money transfers. This method combines easily accessible AI tools, which can generate convincing voice replicas from very short audio samples, with traditional social engineering tactics to enhance the scam's credibility.

OpenAI has scrubbed all mentions of the “io” brand, its planned AI hardware startup with Jony Ive, from its website and social media channels, The Verge says in a report. The company confirmed the move was due to a court order following a trademark complaint from Iyo, a hearing device startup. An OpenAI spokesperson stated that while they disagree with the complaint and are reviewing options, the underlying hardware development deal with Ive's team is still happening. The removed content included an announcement blog post and a video featuring Sam Altman and Jony Ive.