Artificial Intelligence News Live: AI development increasingly focused on emotional intelligence over traditional logic benchmarks

A trend is emerging in the artificial intelligence industry where companies are shifting focus from traditional performance metrics like scientific knowledge and logical reasoning to developing more emotionally intelligent and empathetic language models, according to a report in TechCrunch. This push is driven by competition based on 'soft measures' such as user preference and the perceived helpfulness and naturalness of AI interactions. The development of simulated emotional intelligence is becoming a key differentiator for foundation models as the industry aims to create more sophisticated and human-aligned AI.

A new study examining the privacy practices of various generative AI models has ranked Mistral AI's Le Chat and xAI's Grok as the top performers for respecting user data, according to a report in ZDNet. The research, which assessed factors like data collection policies and user controls, also identified other, unnamed models as being the 'worst offenders' regarding privacy protection. The findings highlight the significant differences in data handling and privacy approaches among competing AI services.

Google has updated its Gemini AI assistant on Android with a new music identification feature, according to a report in ZDNet. The function allows users to ask Gemini to identify a song playing nearby or to recognize a tune from the user humming it. This update integrates Google's pre-existing audio recognition technology, similar to its 'Hum to Search' feature from 2020, directly into the conversational AI interface. The feature is part of a broader strategy to embed Gemini's multimodal capabilities across Google's product ecosystem.

AI medical scribe company Abridge has raised $300 million in a Series E funding round, doubling its valuation to $5.3 billion in just four months, according to a report in TechCrunch. The round was led by Andreessen Horowitz (a16z) and follows a recent $250 million fundraise, highlighting rapid investor interest in the company. Abridge's technology uses generative AI to listen to and summarize patient-doctor conversations, automatically creating structured clinical notes to reduce administrative work for physicians.

Artificial intelligence labs, including Anthropic, Google, and OpenAI, are actively working to overcome the "black box" problem in AI by using "chains-of-thought" to better understand how their models operate, the Financial Times says in a report. This method, which prompts a model to produce step-by-step reasoning, is being used not just to improve performance but as a research tool to interpret the internal decision-making processes of these complex systems. The effort is part of a broader push for mechanistic interpretability, aiming to make AI more transparent, safer, and easier to debug by mapping the models' internal functions to human-understandable concepts.

In a significant copyright case, a federal judge has ruled that AI company Anthropic's practice of training its models on legally purchased physical books that it digitized is considered fair use, according to a report in The Verge. Judge William Alsup of the Northern District of California issued the first-of-its-kind ruling, describing the use of legally acquired content for AI training as "sufficiently transformative." However, the victory for Anthropic is limited, as the judge also ruled that the company must face a separate trial for storing and potentially using millions of pirated books downloaded from the internet, which does not fall under fair use. The decision explicitly does not address the separate legal question of whether the outputs generated by AI models infringe on copyrights.

Google DeepMind has launched a new AI model, Gemini Robotics On-Device, which allows robots to perform tasks locally without needing an internet connection, according to a report in TechCrunch. This specialized model is an optimized version of the broader Gemini AI family, designed specifically for efficient execution on robotic hardware. The system enables robots to translate natural language commands into physical actions and operate autonomously in disconnected environments, which reduces latency and enhances data privacy by keeping processing on the device.

Walmart has developed and deployed its own internal AI platform, called 'Element', which is now used by 1.5 million of its associates, VentureBeat says in a report. Described as an "AI foundry," the platform is designed to build and deploy applications that have reportedly reduced the time associates spend on certain tasks by 67%. The system currently handles approximately three million queries from employees each day. The development of this proprietary platform reflects a strategy to standardize AI deployment, maintain strategic control over its AI applications and data, and increase operational efficiency by augmenting its workforce.

A technology startup, Osmo, is applying artificial intelligence to the fragrance industry, significantly reducing development time for new scents, according to a report in The Verge. The company's system, described as 'artificial olfactory intelligence,' converts scent profiles into digital code that can be manipulated by AI. This process allows Osmo to produce a custom fragrance sample within 48 hours of receiving a client's request. This rapid turnaround contrasts sharply with the traditional perfume creation process, which can take weeks or months for formulation and up to 18 months for a product to go from concept to market.

Identity verification company Persona reports it has blocked 75 million fraudulent job candidates using deepfake technology, highlighting a significant increase in AI-driven hiring fraud, as per a report in VentureBeat. The trend is compelling corporations to deploy more advanced identity verification systems to prevent deepfake infiltration during the hiring process. This surge in fraud is linked to the widespread accessibility of generative AI tools that make creating convincing synthetic media easier, forcing companies to implement stronger security measures for recruitment.

Broadway is currently experiencing a trend of integrating advanced technologies, including artificial intelligence for visual effects and character depiction, into its storytelling, according to a report in the New York Times. Current productions are utilizing high-tech elements such as video projections to display AI-generated actors, visualize the digital memories of robot characters, and create immersive environments. This integration of AI and sophisticated visual effects is creating new creative possibilities for set design and narrative expression, while also raising questions about intellectual property and the future roles of creative professionals in theater.

Robo-advisory firm Wealthfront, which utilizes artificial intelligence for automated financial planning services, has confidentially filed for a US initial public offering, according to a report in Bloomberg. These platforms use AI and algorithms for tasks like portfolio management and risk assessment. The move is seen as an indicator of a potential revival in the financial-technology IPO market following a significant slowdown.

Microsoft is struggling with the enterprise adoption of its Copilot AI assistant because many corporate employees prefer to use OpenAI's ChatGPT, according to a report in Bloomberg. This user preference for the direct, standalone interface of ChatGPT presents a challenge to Microsoft's strategy of deeply integrating Copilot within its software ecosystem like Microsoft 365 and Windows. While Microsoft is OpenAI's largest investor and its Copilot products are powered by OpenAI's models, the situation is creating competitive tension as both companies vie for market share in the enterprise AI sector.

The UK's Competition and Markets Authority (CMA) is intensifying its scrutiny of Alphabet Inc.'s Google, focusing on its online search and advertising business, as per a report in Bloomberg. The investigation centers on the company's new AI-powered search summaries, known as Search Generative Experience (SGE), and concerns that Google could use them to unfairly favor its own services. This regulatory pressure is an extension of the CMA's ongoing examination of Google's market dominance and potential for anti-competitive behavior in the digital advertising space.

Amazon is building one of its largest-ever data centers on a 1,200-acre site in Indiana, a facility supersized to handle the immense computing demands of artificial intelligence partner Anthropic, the New York Times said in a report. This development is part of a broader AI infrastructure arms race among major technology companies to build massive supercomputing clusters for AI model development and cloud services. The project highlights the trend of strategic partnerships between cloud providers and AI startups, and it underscores the escalating energy and resource demands required to power advanced AI systems.

Verizon is overhauling its customer service by integrating Google's Gemini AI into a new "AI-powered Verizon Assistant" within its relaunched My Verizon app, The Verge says in a report. The AI chatbot is designed to manage tasks such as account upgrades, billing questions, and adding new lines, with the goal of improving efficiency. Verizon states the system has shown over 90 percent accuracy and will transfer customers to a human agent if requested or if it cannot resolve an issue. The company is also introducing human "Customer Champions" who will use Google's AI tools for complex problems and is expanding its live agent availability to include 24/7 chat support.

Google is collaborating with the nonprofit Earth Fire Alliance and satellite company Muon Space to improve the detection of wildfires using satellite imagery and artificial intelligence, according to a report in Wired. The partnership aims to leverage Google's AI capabilities, specifically computer vision, to process large amounts of data collected by Muon Space's satellites. This initiative seeks to provide faster and more accurate tracking of fires by automating the analysis of satellite imagery for early signs of wildfires, such as heat signatures and smoke plumes, thereby improving upon existing detection methods.

Google has introduced its AI-powered search feature, described as an "AI mode," to users in India, according to a report in TechCrunch. This new question-and-answer style tool is currently in an experimental phase and requires users to opt-in via Search Labs to access it. After opting in, users can utilize the feature by asking queries in English. The company has not yet provided details on whether support for other languages will be added in the future.

A new study from the University of Amsterdam reveals that the human brain automatically processes 'action possibilities,' or affordances, in an environment, a form of intuitive physical understanding that current AI systems like ChatGPT lack, according to a report in ScienceDaily. Researchers found that specific brain regions activate in response to these potential actions, such as swimming or walking, independent of direct visual cues. This contrasts sharply with current AI models, which struggle with such intuitive judgments because they lack the physical context and embodied experience that humans naturally possess.

Researchers at MIT have developed a framework known as SEAL (Self-supervised Editing for Adaptable Language models) designed to enable large language models to continuously learn new information and tasks, as per a report in VentureBeat. This approach addresses the significant challenge of "catastrophic forgetting," where AI models lose previously acquired knowledge upon learning new things. The SEAL framework aims to overcome the limitations of current static AI models, which become outdated after their initial training and require computationally expensive retraining to update their knowledge base. By allowing models to autonomously adapt and integrate new data, this method facilitates the development of more dynamic and persistently relevant AI systems.

The U.S. National Highway Traffic Safety Administration (NHTSA) has contacted Tesla following the emergence of multiple online videos that appear to show the company's AI-driven robotaxis violating traffic laws, according to a report in TechCrunch. The regulatory inquiry is focused on the vehicles' performance in South Austin, where Tesla is currently operating a limited, invitation-only service. This action by the federal safety agency represents an initial information-gathering step and is not yet a formal investigation or finding of fault.

Meta Platforms CEO Mark Zuckerberg discussed a possible acquisition with generative AI video startup Runway AI Inc., according to a report in Bloomberg. The discussions were reportedly part of a broader deal-making and recruiting drive by Meta to bolster its artificial intelligence capabilities. The talks indicate a strategic interest in acquiring advanced technology and specialized talent in the competitive generative AI sector, specifically focusing on video synthesis.

In a court filing related to his ongoing lawsuit against OpenAI and its CEO Sam Altman, lawyers for Elon Musk have claimed he 'does not use a computer,' according to a report in Wired. The assertion was made within the legal proceedings where Musk alleges the AI company has abandoned its original non-profit mission. This legal claim is noted to be in contrast with Musk's public record, which includes multiple posts on social media platforms about his laptop usage.