Latest News

A major update has been made to the way developers interact with Google’s Gemini models and agents. The Interactions API, which was launched in public beta last December, is now generally available and serves as the primary interface for these applications. This shift marks a significant change from the legacy generateContent API, although the latter will continue to be supported for the foreseeable future.

The Interactions API offers several key features that make it an attractive choice for developers building with Gemini models and agents. These include server-side state, background execution, tool combination, and multimodal generation capabilities. With this release, a stable schema has been established, and major new functionalities have been added in response to developer requests.

One of the most significant additions is Managed Agents, which allows for more efficient management of agent resources. Background execution enables long-running tasks to be performed without blocking other operations. Gemini Omni will also soon be integrated into the API. These features are designed to support stateful and agentic workflows, making them particularly well-suited for applications that require persistent interactions with users or environments.

The Interactions API is now the default interface across various Google AI tools, including Google AI Studio and the Gemini API. All documentation has been updated to reflect this change, although a toggle remains available to switch back to legacy format snippets. Developers are encouraged to use the Interactions API for new projects and applications, as it offers improved flexibility and efficiency.

For developers already building with supported partners such as LiteLLM, Eigent, or Agno, integrations for the Interactions API are readily available through these platforms. The gemini-interactions-api Skill has also been developed to inject best-practice patterns into agent development contexts, making it easier for agents to stay up-to-date with the latest API developments.

The Interactions API is accessible via Python and JavaScript SDKs. To get started, developers can obtain an API key from Google AI Studio and consult the comprehensive documentation provided by Google. A migration guide has been published to assist those transitioning from the legacy generateContent API, which maps every field to the new schema for a seamless transition.

Google continues to prioritize developer feedback in shaping the Interactions API’s future development. The company encourages developers to share their needs and suggestions on the developer forum, ensuring that the API remains responsive to its users’ requirements.

Tesla has finally started rolling out its latest Full Self-Driving (FSD) stack in Australia and New Zealand. However, owners of the Model Y Long Range (L) variant will have to wait a bit longer for the update. According to an official email notification shared on social media by @nathanhome, Tesla is still working on refining FSD v14 for this specific model before making it available more broadly.

In the first quarter of this year, Tesla sales in the United States took a significant hit, marking its third consecutive year of quarterly declines. The automaker sold 117,300 units between January and March, representing its lowest quarterly sales since late 2021 and an 8.4% year-over-year decline, according to estimates from Cox Automotive. This downturn is part of a broader trend in the U.S. electric vehicle (EV) market, which saw total EV sales plummet by 27% compared to the same period last year.

The data analysis tools used by InsideEVs reveal that non-Tesla EV sales fell even harder, with a staggering 41% decline. This collapse of demand has allowed Tesla to maintain its position as the leading U.S. EV seller and even increase its market share from 43.2% to 54.2% year over year.

The expiration of federal tax credits last September and the end of regulations driving clean-car sales have had a profound impact on the industry, forcing automakers to reassess their electric vehicle ambitions. Many companies are scaling back their EV plans, reallocating resources towards combustion engine models and hybrids as they cancel or put on hold various projects.

Tesla is not immune to these headwinds, but its core passenger-vehicle business has taken a back seat under Elon Musk’s pivot toward AI and robotics. The company hasn’t released a new model since the Cybertruck in late 2023, which was met with lukewarm reception.

The Model Y remains Tesla’s top seller, accounting for one-third of all EVs sold in Q1 and making up 67% of its own sales. Its competitive pricing and strong range have kept demand high despite the broader market contraction. In fact, Model Y sales grew by 22% year over year, according to Cox estimates.

However, other Tesla models are struggling. The Model 3 posted one of its worst quarters in years, with an estimated 31,672 sales – a nearly 40% decline from last year’s Q1 results. This is particularly concerning given the preference for electric SUVs and crossovers among American buyers, which has left sedans like the Model 3 struggling to compete.

The Cybertruck’s performance was even more dismal, with a 45% year-over-year drop in sales and just 3,513 units sold. Despite initial hype, this wedge-shaped stainless steel truck has turned out to be an epic flop, with only a dedicated group of fans keeping it afloat. Sales of the Model S and Model X, which have been discontinued, also fell sharply but haven’t contributed significantly to Tesla’s sales for some time.

Looking ahead, there are hints that new models may be in the pipeline, including a compact SUV reportedly cheaper and smaller than the Model Y. Elon Musk has even teased ‘something way cooler than a minivan’ on his X account, fueling speculation about future releases. However, until these plans materialize, Tesla will continue to rely heavily on its current offerings – particularly the Model Y.

For now, this single model is doing all the heavy-lifting for the brand, maintaining market share and even growing it despite the industry’s decline. But as electric car sales grow over time, Tesla could find itself outflanked by competitors who have been more proactive in adapting to changing consumer preferences.

Fire detection is crucial in subway tunnels, where confined spaces and ventilation systems can complicate safety monitoring. Existing approaches often rely on classical machine learning methods that detect fires based on multivariate correlations but struggle with contextual reasoning.

However, recent studies suggest that Large Language Models (LLMs) may help address this challenge by translating structured sensor summaries into concise semantic descriptions. To examine the role of LLMs in fire detection, researchers have proposed a hybrid framework called HyFiD, which employs an LLM as a semantic feature extractor to augment classical classifiers.

HyFiD converts momentary multi-sensor readings (temperature, smoke, O2, CO, and CO2) into textual assessments of environmental states, generating semantic vectors that are fused with numerical features for supervised classifier training. Experiments on Fire Dynamics Simulator (FDS)-based subway scenarios show that the effect of LLM-derived semantic features is strongly dependent on the downstream classifier.

The results indicate a clear trade-off between detection sensitivity and premature-alarm suppression. Lower thresholds generally preserve higher F1 scores but substantially increase Pre-Alarm Rate, indicating more aggressive alarm behavior. For instance, at τ = 0.25 and k = 1.0 s, SVM + LLM and GBM + LLM achieve high F1 scores of 87.45% and 90.85%, respectively, but their PAR values rise to 99.25% and 51.32%. In contrast, higher thresholds suppress premature alarms more strongly but often degrade F1.

The temporal persistence window further modulates this trade-off. Increasing the persistence window from 1.0 to 2.0 s or 3.0 s generally reduces PAR, but excessive smoothing can substantially reduce F1, especially for SVM + LLM and RF + LLM. GBM + LLM is comparatively more robust under moderate smoothing: at τ = 0.50 and k = 2.0 s, it maintains an F1 score of 88.33% while reducing PAR from 47.55% under the default setting to 28.68%. However, further increasing the window to k = 3.0 s reduces F1 to 83.46%, showing that overly conservative alarm persistence can also suppress valid fire detections.

To justify the architectural design of using the LLM as a semantic feature extractor rather than as an end-to-end decision maker, researchers evaluated direct LLM classification under Zero-shot, One-shot, and Few-shot Chain-of-Thought (CoT) prompting strategies. The results show that direct prompting is highly sensitive to both the prompting setting and the LLM backbone when applied to raw multivariate numerical sensor snapshots.

The final alarm behavior is highly sensitive to the selected threshold and persistence window. Researchers retain τ = 0.50 and k = 1.0 s as the default setting for the main experiments because it provides a consistent operating point for comparing classifiers, while the sensitivity analysis clarifies how alternative alarm rules shift the balance between F1, delay, and pre-ignition false alarms.

Under the Zero-shot CoT setting, several evaluated backbones—including Llama-3-8B41, Qwen-2.5-3B58, Qwen-2.5-7B, and Ministral-8B59—produce near-zero Recall and F1, indicating difficulty in directly mapping raw numerical sensor values to stable fire/no-fire decisions. However, Few-shot CoT improves the strongest prompt-only result, with Llama-3-8B reaching an Accuracy of 65.67% and an F1 score of 70.32%. Nevertheless, this remains below the supervised hybrid approach: GBM + Llama-3-8B achieves an F1 score of 90.77% when the LLM is used as a semantic feature generator rather than as a standalone classifier.

These comparisons suggest that, in this setting, LLMs are more reliable when used to produce intermediate semantic representations for supervised downstream classifiers than when used as direct numerical classifiers. This finding has significant implications for machine learning jobs and AI-generated image applications, where accurate decision-making is critical.

New Taipei City, Taiwan-based Vecow Co., Ltd. has announced its participation in the recent AUTOMATE 2026 event held in Chicago. The company showcased its latest AI robotics platforms and deployment-ready solutions designed to accelerate industrial automation workflows.

San Diego-based Qualcomm Technologies has announced the selection of fifteen startups from Japan, Singapore, and South Korea to participate in its AI Program for Innovators (QAIPI) 2026. The program aims to empower these startups with cutting-edge technologies and resources to develop scalable edge AI solutions that address real-world challenges across various industries.

The integration of generative models like ChatGPT, Claude, and Gemini into daily work has brought about a pressing question in the content creation industry: how to maintain authentic voice. As professionals, students, and creators rely on these tools for assistance, humanization becomes increasingly important. AI Humanizer Tools have emerged as a solution to this problem, rewriting machine-generated text into language that reads naturally and avoids formulaic patterns.

Samsung Electronics is accelerating its adoption of artificial intelligence (AI) across the company by deploying ChatGPT Enterprise and Codex to its employees worldwide. The deployment, one of OpenAI's largest enterprise deployments to date, will make these tools available to all Samsung Electronics employees in Korea and those working in its Device eXperience (DX) division globally.

Epson Robots is set to preview its upcoming collaborative robot platform and demonstrate SafeSense technology at the upcoming Automate 2026 event. The exhibition, being held from June 22 to 25 at McCormick Place in Chicago, will give attendees their first look at Epson's latest innovation in automation solutions for plastics manufacturers.

Chicago will be the hub of industrial automation innovation this June as the Association for Advancing Automation (A3) hosts its annual conference, Automate 2026. The event brings together robotics, machine vision, motion control, and industrial AI communities to showcase the latest advancements in these fields.

Creating professional music visuals no longer requires a studio budget or a motion graphics team. With the right tools, independent artists and producers can turn raw audio into engaging visual content without breaking the bank. Whether you're releasing your first single or building a visual brand, these seven essential tools offer varying levels of affordability, customization, and ease of use to suit different creative needs.

Glean, a company that specializes in developing artificial intelligence (AI) assistants, has announced an expansion of its financial services ecosystem. The move is designed to help regulated industries such as banking and finance deploy AI agents and build workflows that address real operational needs. This initiative comes at a critical time for the industry, where firms are under pressure to improve productivity and compliance while managing data privacy and reliability risks.

Tesla has quietly rolled out a series of updates to its flagship Model S vehicle, introducing a range of new features designed to enhance the driving experience. Among these changes is a high-visibility glass roof that allows for significantly more light to enter the cabin, providing an unobstructed view of the sky day or night. This innovative design not only improves visibility but also reduces weight compared to its predecessor while maintaining heat and UV protection.

A major expansion of Adobe's plan to integrate artificial intelligence (AI) into its Creative Cloud suite is now underway, with the company rolling out bespoke AI assistants to its most popular editing and design apps. The new chatbots are designed to automate tasks and provide a more intuitive way for users to work within Photoshop, Premiere, Illustrator, InDesign, and Frame.io.

Google DeepMind has released a comprehensive report titled 'From AGI to ASI,' written by Tim Genewein and 13 coauthors. The report explores four main technical pathways from Artificial General Intelligence (AGI) to Artificial Superintelligence (ASI), analyzing the potential bottlenecks that could slow or shape these paths.