AI News

At this event, we'll explore self-awareness in LLMs, as described in the paper [Tell me about yourself: LLMs are aware of their learned behaviors]. Guiding us through the topic will be one of the paper's co-authors, Jenny Bao.

Enterprise artificial intelligence investment is unprecedented, with IDC projecting global spending on AI and GenAI to double to $631 billion by 2028. Yet beneath the impressive budget allocations and boardroom enthusiasm lies a troubling reality: most organisations struggle to translate their AI ambitions into operational success.

Meta has introduced its first AI-powered video editing feature, now available in the Meta AI app, on the Meta.AI website, and in the new Edits app in the US and over a dozen other countries, with more than 50 preset styles for users to choose from.

Sam Altman's essay 'The Dream of a Gentle Singularity' is analyzed. The author questions the optimistic predictions about the ease with which we will adapt to superintelligence and the plan to 'solve alignment'. The author highlights the lack of concrete details on solving the alignment problem and expresses skepticism about the society's ability to navigate the changes that will occur, and its unwillingness to regulate the industry.

This blog post describes how to post-train (fine-tune) GR00T N1.5 using teleoperation data from a single SO-101 robot arm. GR00T N1.5 can be easily fine-tuned and adapted using the affordable, open-source LeRobot SO-101 arm. The tutorial explains how to prepare the dataset, fine-tune the model, evaluate the policy, and deploy it onto a physical robot.

Generative AI is reshaping the way people work, from full-time employees to freelancers. As coding copilots, design assistants, and AI-powered writing tools become more capable and accessible, creative and technical roles are starting to shift – if not become eliminated entirely. The pressure to adapt is growing across the board.

Pinterest is experimenting with “auto-collages,” an AI feature that allows advertisers to quickly turn their product catalogs into shoppable collages.

Gaining visibility into enterprise cloud assets is growing ever more challenging. CloudQuery is addressing this by centralizing cloud assets, security metadata and cost in one place, accessible through SQL queries. The company is announcing a $16 million funding round led by Partech.

As artificial intelligence reshapes everything from education to defense, digital sovereignty is no longer a long-term aspiration; it's an urgent priority. To ensure that nations can govern, audit, and shape the AI systems that influence their societies, open source AI offers a powerful foundation for autonomy, innovation, and trust.

Uptime Industries is taking a bet on on-premise AI infrastructure with an “AI-in-a-box” handheld device called Lemony AI.

Reinforcement learning (RL) (e.g., GRPO) helps with grounding because of its inherent objective alignment—rewarding any successful clicks—rather than encouraging long textual Chain-of-Thought (CoT) reasoning (also referred to as “thinking”). In this blog, we will share a complete recipe towards training state-of-the-art GUI grounding models using GRPO.

OpenAI chief Sam Altman has declared that humanity has crossed into the era of artificial superintelligence—and there’s no turning back.

Mistral Compute is a new AI infrastructure offering that will provide customers a private, integrated stack—GPUs, orchestration, APIs, products, and services in whatever form factor they need, from bare-metal servers to fully-managed PaaS.

On Wednesday, Disney and NBCUniversal filed a lawsuit against AI image-synthesis company Midjourney, accusing the company of copyright infringement. The complaint, filed in US District Court in Los Angeles, marks the first major legal action by Hollywood studios against a generative AI company. The studios claim Midjourney downloaded copyrighted content using “bots, scrapers, streamrippers, video downloaders, and web crawlers” to train its AI model. The lawsuit describes San Francisco-based Midjourney as a "bottomless pit of plagiarism" that enables users to generate what the studios call "AI slop”—personalized images of copyrighted characters.

New studies reveal pattern-matching reality behind the AI industry's reasoning claims.

Today at GTC Paris, Hugging Face and NVIDIA announced Training Cluster as a Service, making large GPU clusters more accessible for AI research. This collaboration provides researchers with the flexibility to pay only for the duration of training runs.

MarketsandMarkets values the global artificial intelligence market at $371.71 billion and expects it to exceed $2407.02 billion in value by 2032. The statistic clearly demonstrates how AI technology can affect many sectors, including cryptocurrency. The Business Research Company reports the generative AI market in the cryptocurrency space alone is expected to grow in value from $760 million in 2024 to $1.02 billion in 2025. That’s a CAGR of roughly 34.5%. As readers will know, artificial intelligence boasts an unusual computational ability that helps it extract meaningful insights in real time.

Teachers in England have been given the all-clear to use AI to help them in low-level tasks. Guidance from the Department for Education (DfE) says AI can be used by school teachers in England, but it should only be for ‘low stakes’ tasks, such as writing letters to parents and marking homework.

In a romanticized essay on the supposedly imminent singularity, OpenAI CEO Sam Altman mentions an interesting fact about ChatGPT's energy consumption: a single ChatGPT request consumes an average of 0.34 watt-hours of power and 0.000085 gallons of water. That energy usage is roughly on par with what a Google search consumed back in 2009, but Altman excludes the fact that ChatGPT is likely handling far more requests per person. And with new AI models like multimodal systems, agents, and advanced reasoning engines demanding even more compute, the rapid expansion of data centers suggests the overall energy appetite of these systems is only increasing.

OpenAI is delaying its first open-weight language model since GPT-2 until later this summer, CEO Sam Altman said on X. Originally planned for release before the end of June, the model will include reasoning capabilities. Altman said the research team made unexpected progress that now requires more time, calling the result "very worth the wait."