Notizie — Pagina109

Amazon Web Services unveiled a trio of upgrades to its Bedrock AgentCore platform at the re:Invent conference. The new Policy tool lets developers set natural‑language boundaries for AI agents, including data‑access limits and transaction caps, while integrating with the AgentCore Gateway to enforce those rules. A pre‑built Evaluations suite offers 13 metrics to monitor correctness, safety and tool‑selection accuracy. Finally, AgentCore Memory gives agents a persistent log of user preferences, enabling more personalized interactions. Vice‑president David Richardson emphasized that these layers together aim to make enterprise AI agents safer, more controllable, and more useful over time.Leggi di più

Simular, a startup developing AI agents that can directly control Mac OS and Windows computers, announced a $21.5 million Series A round led by Felicis with participation from NVentures and other investors. The company released version 1.0 of its Mac agent and is collaborating with Microsoft on a Windows version as part of the Windows 365 for Agents program. Founded by former Google DeepMind researchers Ang Li and Jiachen Yang, Simular’s technology, dubbed “neuro‑symbolic computer‑use agents,” aims to curb LLM hallucinations by generating deterministic code for repeatable tasks. Early customers include a car dealership and homeowners’ associations.Leggi di più

French AI startup Mistral unveiled its Mistral 3 family, featuring a large frontier model with multimodal and multilingual capabilities and nine smaller, fully customizable models. The launch emphasizes open-weight access, allowing developers to run models on a single GPU and fine‑tune them for specific enterprise tasks. Mistral positions its models as cost‑effective alternatives to closed‑source rivals, highlighting efficient architecture, extensive context windows, and suitability for on‑premise deployment. The company also announced collaborations with partners in robotics, cybersecurity, and automotive sectors to integrate its models into specialized applications.Leggi di più

French AI firm Mistral AI announced a new family of open‑source language models, including the flagship Mistral Large 3 and a suite of smaller Ministral 3 variants. Designed for general‑purpose use and edge deployment, the models support a wide range of languages by emphasizing non‑English training data. Co‑founder Guillaume Lample highlighted the goal of making high‑end AI accessible, while the open‑weight approach allows developers to customize and run the models on laptops, smartphones, cars, or robots without relying on constant internet connectivity.Leggi di più

OpenAI CEO Sam Altman announced a internal "code red" for ChatGPT, pausing work on new initiatives to concentrate on speed, reliability and personalization. The move comes as competitors such as Google’s Gemini app gain traction, prompting OpenAI to prioritize user experience over experimental features and monetization plans.Leggi di più

Anthropic’s societal impacts team, led by Deep Ganguli, examines how the company’s Claude chatbot is used and how it influences society. The small group of researchers and engineers gathers usage data through an internal tool called Clio, publishes findings on bias, misuse, and economic impact, and works closely with safety and policy teams. Their work includes identifying explicit content generation, coordinated spam, and emerging emotional‑intelligence concerns such as “AI psychosis.” While the team enjoys a collaborative culture and executive support, it faces resource constraints as its scope expands.Leggi di più

Gradium, a Paris‑based AI voice startup spun out of the French lab Kyutai, announced a $70 million seed financing led by FirstMark Capital and Eurazeo, with participation from Xavier Niel, DST Global Partners and Eric Schmidt. The company, founded by former Google DeepMind researcher Neil Zeghidour, offers ultra‑low‑latency, multilingual voice models that aim to deliver near‑instantaneous AI speech. Gradium enters a crowded market that includes major LLM firms and specialized voice startups, positioning its technology for developers seeking faster, more accurate voice capabilities across multiple languages.Leggi di più

A joint study by MIT, Northeastern University and Meta reveals that large language models can rely heavily on sentence structure, sometimes answering correctly even when the words are nonsensical. By testing prompts that preserve grammatical patterns but replace key terms, the researchers demonstrated that models often match syntax to learned responses, highlighting a potential weakness in semantic understanding. The findings shed light on why certain prompt‑injection techniques succeed and suggest avenues for improving model robustness. The team plans to present the work at an upcoming AI conference.Leggi di più

Chinese startup DeepSeek has released two new AI models—DeepSeek‑V3.2 and DeepSeek‑V3.2‑Speciale—under an open-source license. The models claim performance comparable to GPT‑5 and Gemini 3 Pro on long‑form reasoning, tool use, and dense problem solving while offering a 128,000‑token context window and reduced computational cost through Sparse Attention. Their launch challenges the dominance of U.S. AI firms, sparks regulatory scrutiny in Europe, and raises questions about the future of AI accessibility and geopolitics.Leggi di più

OpenAI chief executive Sam Altman announced an internal “code red,” pausing projects such as ads, shopping, health agents, and the Pulse personal assistant to focus on boosting ChatGPT’s speed, reliability and personalization. The memo calls for daily calls and temporary team transfers to speed development. Meanwhile, Google, which launched its own “code red” after ChatGPT’s debut, sees its AI user base expand with tools like Nano Banana and its new Gemini 3 model, which outperforms rivals on several benchmarks. The parallel moves highlight a pivotal moment in the AI race, with both firms investing heavily to maintain leadership.Leggi di più

ChatGPT is a powerful tool, but it isn’t suitable for every task. Experts warn against relying on the AI for diagnosing health conditions, mental‑health support, emergency safety decisions, personalized financial or tax advice, handling confidential data, illegal activities, academic cheating, real‑time news monitoring, gambling, drafting legal contracts, or creating art to pass off as original. While it can help with general information and brainstorming, users should treat it as a supplement, not a replacement for professional expertise or critical real‑time resources.Leggi di più

A new BloombergNEF report projects that data centers will need nearly three times the electricity they consume today, rising to 106 gigawatts by 2035. Growth will be driven by larger facilities, higher utilization rates and the surge in AI training and inference workloads. Much of the new capacity is expected in rural regions across the PJM Interconnection and Texas’s ERCOT grid, prompting regulatory scrutiny over grid reliability and electricity pricing.Leggi di più

Apple announced that John Giannandrea, the company's AI chief since 2018, is stepping down and will serve as an adviser through the spring. He will be succeeded by Amar Subramanya, a veteran who spent 16 years at Google before leading engineering for the Gemini Assistant at Microsoft. The leadership change comes as Apple Intelligence has faced a series of setbacks, including missteps with Siri and inaccurate news summaries. Subramanya’s deep experience with rival platforms is expected to help Apple address its AI challenges and accelerate development of on‑device AI services.Leggi di più

Director James Cameron warned that AI‑generated actors are "horrifying," expressing concern that synthetic performers could replace real talent. The comment followed the debut of Tilly Norwood, a photorealistic digital actress created by Particle6 and shown at the Zurich Film Festival. SAG‑AFTRA condemned the technology as a synthetic imitation built on stolen work. Cameron, known for pioneering CGI, distinguished motion‑capture—which still relies on human performers—from generative AI that can fabricate characters and performances from text prompts. He urged the industry to keep the human element at the core of filmmaking.Leggi di più

OpenAI announced an ownership stake in private‑equity firm Thrive Holdings, a move that involves no cash outlay but provides Thrive’s portfolio companies with OpenAI employees, models, products, and services. The partnership focuses on transforming high‑volume, rules‑driven processes in IT services and accounting, aiming to boost speed, accuracy, and cost efficiency. Thrive CEO Joshua Kushner highlighted AI’s potential to reshape industries from the inside out, while OpenAI’s leadership described the deal as a new model for collaboration with private‑equity groups. The arrangement also gives OpenAI access to data that could enhance future AI training.Leggi di più

Apple announced that its head of artificial intelligence, John Giannandrea, will leave his role, with Amar Subramanya set to take over as vice president of AI. The change comes as Apple faces setbacks with its Siri voice assistant, which has been delayed and reportedly caused confidence concerns among senior leadership. Giannandrea will remain as an advisor before retiring, while Subramanya, a former Google veteran, will oversee AI models, research, and safety, reporting to software chief Craig Federighi.Leggi di più

OpenAI faces pressure to reveal why it removed two internal datasets built from a shadow library of pirated books. The move comes amid a class‑action lawsuit from authors who allege the company trained ChatGPT on their works without permission. While OpenAI initially said the datasets were deleted because they fell out of use, it later claimed that any reason for deletion is protected by attorney‑client privilege. A U.S. district judge has ordered the company to produce internal communications about the deletion, including references to the library source.Leggi di più

DeepSeek, the Hangzhou‑based artificial‑intelligence firm, announced the release of two new reasoning‑capable models, V3.2 and V3.2‑Speciale. V3.2 is now available through the company’s app and web platform, while V3.2‑Speciale is offered via API access only. The company claims the Speciale version surpasses Google’s Gemini 3.0 Pro and OpenAI’s GPT‑5 High on benchmark tests, positioning DeepSeek as a serious contender in the global AI race.Leggi di più

Nvidia announced a new open‑source vision‑language model called Alpamayo‑R1 at the NeurIPS AI conference. Designed for autonomous‑driving research, the model builds on Nvidia’s Cosmos‑Reason architecture and aims to give self‑driving systems common‑sense reasoning. Nvidia also released a set of developer guides known as the Cosmos Cookbook, and made the model available on GitHub and Hugging Face. Executives highlighted the importance of physical AI for the next wave of robotics and autonomous technologies.Leggi di più

OpenAI announced an ownership stake in Thrive Holdings, a private‑equity‑style firm backed by Thrive Capital that aggregates AI‑focused companies. While the deal’s financial terms remain confidential, OpenAI will embed its engineering, research, and product teams within Thrive’s portfolio to accelerate AI adoption. The partnership follows OpenAI’s recent circular investments in infrastructure firms such as Advanced Micro Devices and CoreWeave, underscoring a broader strategy of aligning with companies that can embed its technology. Analysts say the move will be closely watched to gauge whether the combined entities can generate sustainable growth or simply inflate valuations.Leggi di più