Search results for "gemini Google":

Google announced a new "Personal Intelligence" feature for its Gemini robot. This feature allows Gemini to securely connect and access information from users' Gmail, Google Photos, YouTube, and other applications, providing assistance more tailored to individual user needs through cross-source reasoning. Currently, this feature is in beta, initially available to Google AI Pro and AI Ultra subscribers in the United States. Cross-application reasoning: Finding license plates and planning trips no longer requires rummaging through files. Google emphasizes that the core advantages of "Personal Intelligence" lie in "cross-source reasoning" and "detailed extraction." After users actively enable the connection in their settings, Gemini can combine text, photos, and video information to answer questions. Google gave a very practical example: when a user needs to change the tires on their 2019 Honda SUV, Gemini can not only provide tire specifications like a regular robot, but also delve into Google Photos to analyze vehicle photos and suggest tire models suitable for local road conditions. Even when users need license plate numbers, Gemini can directly identify the license plate from the photo album and confirm the specific car model by searching Gmail, saving users the trouble of switching between different apps. Furthermore, in travel planning, Gemini can analyze users' past Gmail and photo album records to understand family members' interests, thus avoiding crowded tourist spots and recommending itineraries and board games more suited to family preferences. Privacy Design: Off by Default, Not Directly Used for Model Training Regarding users' most pressing privacy concerns, Google states that this feature is "off by default," giving users complete control over which applications to connect to. Google emphasizes that Gemini accesses data to respond to requests, but "does not directly use users' Gmail inboxes or Google Photos libraries to train models." Google only uses filtered, obfuscated personal data with "specific prompts" and "model responses" for subsequent optimization training. The system is trained to "understand how to find information," not to "remember your information." This is still in beta: Beware of AI "over-interpreting" While powerful, Google admits it's still in testing and may experience "over-personalized" misjudgments. For example, if a user's album contains hundreds of photos of them at golf courses, Gemini might mistakenly assume the user "loves golf," ignoring the fact that the user was simply accompanying their son and may not actually enjoy the sport. Google urges users to correct such misjudgments directly in the conversation (e.g., "I don't like golf") or provide negative feedback to help the model adjust its understanding of interpersonal relationships and interests. Launch Information: "Personalized AI Services" will be rolled out to eligible Google AI Pro and AI Ultra subscribers in the US over the next week (excluding Workspace business/education versions). The feature supports web, Android, and iOS, and will be expanded to Google Search's AI mode and more countries and regions in the future. Analysis...

While Apple and Google confirmed in an earlier joint statement that they would collaborate in the field of AI, making the Gemini model the cornerstone of Apple's next-generation AI capabilities, the official statement downplayed the technical details. The Information, however, released an in-depth report revealing many interesting details behind this collaboration, including Apple's high degree of autonomy, with even Siri's answers appearing without any Google influence. Although using Gemini, it doesn't look like Google at all. According to The Information, citing sources familiar with the matter, this collaboration is not simply an "API integration." Apple has gained significant control in the contract, the most crucial aspect being "independent fine-tuning." This means that although the underlying model is Google's Gemini, Apple can independently fine-tune it, deciding how the model should answer user questions. Apple can request Google to adjust certain parameters, but the final "tone" and "response logic" will be handled by Apple itself. More importantly, there is "brand invisibility." In current internal prototype testing at Apple, the AI ​​responses driven by Gemini show absolutely no Google or Gemini branding, confirming Bloomberg's previous prediction that Apple wants users to feel like it's still "Siri," not "Google Assistant living in the iPhone." Will Siri be more than just "searching the web"; it can also offer emotional support? The report indicates that Apple hopes Gemini will help Siri shed its habit of simply listing web links. Future Siri will be able to provide complete answers to questions about global knowledge (such as a country's population or scientific principles). Apple also seems to intend to give Siri "emotional support." Sources reveal that Siri previously struggled to handle users expressing "loneliness" or "frustration," but with Gemini, Siri will be able to provide more empathetic and complete conversational responses, similar to ChatGPT. This is undoubtedly a risky move, as there are already numerous cases showing that vulnerable users may become overly reliant on AI chatbots, and the consequences could be disastrous if the AI ​​develops illusions or misinterprets situations. Therefore, how Apple strikes the balance between "thoughtfulness" and "security" will be a key focus of observation going forward. Hybrid Architecture: Handle Small Tasks In-House, Use Gemini for Big Tasks. Regarding the technical architecture, this report echoes the challenges mentioned by Apple's software chief, Craig Federighi. Apple's final solution is to seamlessly integrate "traditional Siri" and "LLM Siri," somewhat similar to the collaboration model between Samsung and Google. • Simple Tasks (On-device): Setting timers, creating reminders, sending simple text messages, etc., are still handled by existing technology on Apple devices, ensuring speed and privacy. • Complex Tasks (Cloud/Gemini): When instructions are ambiguous or require reasoning, Gemini handles the task. For example, if you ask Siri to "text your mother," but there's no contact named "mother" in your address book, the new Siri can analyze your past messages through Gemini to infer which contact is most likely your mother and execute the command. This "reasoning ability" is precisely what Siri lacked in the past. Timeline: Warm-up in spring, WWDC...

After months of rumors and speculation, Apple and Google have finally issued a joint statement officially confirming a multi-year collaboration agreement. Apple will use Google's Gemini model and cloud technology as the core of its Foundation Models, driving a more personalized new version of Siri, expected to debut later this year, as well as other Apple Intelligence features. To save Siri, Apple has finally chosen Gemini. This statement confirms recent reports from CNBC and Bloomberg News. Apple stated that after "careful evaluation," it determined that Google's AI technology provides the "most capable foundation" for its foundation models. This means that the new version of Siri, expected later this year (possibly with iOS 26.4 or at WWDC 2026), will no longer rely solely on Apple's own AI model, but will instead incorporate Google Gemini's computing power and architecture. This collaboration will address Siri's long-standing criticisms of being unintelligent and lacking in comprehension, and unlock more innovative user experiences. Privacy Remains the Bottom Line: Private Cloud Compute Mechanism to Oversee Processing. Despite incorporating Google technology, Apple strongly emphasizes its commitment to privacy. The statement indicates that Apple Intelligence-related functions will continue to run on Apple devices and its own Private Cloud Compute. In other words, Apple uses Gemini as a foundational "knowledge base" or "inference engine," but data processing and privacy protection are still handled and managed through Apple's own architecture, ensuring that users' personal data does not flow directly into Google servers without processing. From Competition to Symbiosis: A $1 Billion Compromise? This collaboration is seen by the market as a symbol of Apple acknowledging its "temporary lag" in the AI ​​race. Back in March 2025, Apple announced a delay in the release of the new Siri, indicating that its self-developed model was not progressing as expected. According to previous market rumors, Apple might have to pay Google up to $1 billion annually in licensing fees for this collaboration. However, this collaboration may be a win-win situation for both Apple and Google. Google is using this to solidify its leading position in generative AI, while gaining access to massive inference demand from billions of iPhones. Apple, on the other hand, can quickly catch up in terms of AI experience, avoiding falling behind in competition with companies like Samsung and OpenAI. Analysis...

At CES 2026, Boston Dynamics announced a strategic partnership with its former employer, Google DeepMind. This collaboration will integrate Google DeepMind's advanced Gemini Robotics model into Boston Dynamics' next-generation Atlas humanoid robot. The two companies plan to launch a joint research program in the coming months, aiming to create the ultimate robot with both the most powerful "cerebellum" (motor control) and the most powerful "brain" (AI reasoning). When the "most powerful body" meets the "most powerful brain," Boston Dynamics is renowned for its robots' amazing balance and agility, and possesses the industry's most advanced robot motion control technology. Google, on the other hand, continues to lead in the field of multimodal AI models, with its Gemini Robotics model specifically designed for various robots, giving them the ability to perceive their environment, reason, use tools, and interact with humans. The next-generation Atlas humanoid robot, unveiled at CES this week, boasts impressive hardware specifications. Equipped with a robotic hand the size of a human, and featuring rotating joints and tactile sensing capabilities, it can lift objects weighing up to 110 pounds (approximately 50 kilograms). Through this collaboration, Atlas will not only move flexibly but also think intelligently. Hyundai Motor Group, the current parent company of Boston Dynamics, is the driving force behind this collaboration, which aims to deploy Atlas, equipped with the Gemini brain, in its factories for testing this year, with the goal of officially deploying it on production lines by 2028. It will be responsible for tasks requiring judgment and dexterous hand movements, such as organizing parts.

Google originally had ambitious plans to have its generative AI Gemini fully replace the old Google Assistant by the end of 2025, becoming the default digital assistant on most Android phones. However, this plan seems to have encountered some real challenges. According to an earlier statement from Google, the timeline has been adjusted, confirming a slowdown in the full rollout of Gemini. Google stated that to ensure a "seamless transition," the original aggressive timeline has been postponed, meaning the update process for fully transitioning from Google Assistant to Gemini will continue until 2026. To avoid compromising the user experience, Google admitted in its statement that it needs more time to optimize and announced that it will share more details in the "coming months," suggesting that the completion date of the full transition may be delayed beyond early 2026. In fact, the "retirement" of Google Assistant was decided long ago. Since Google launched Gemini and began gradually migrating functions originally belonging to Google Assistant (such as controlling smart home devices and setting alarms) to Gemini, the handover was only a matter of time. In the Pixel 9 series phones launched in 2024, Gemini will effectively be the default digital assistant service (although users can still manually switch to Google Assistant). Expanding to tablets and wearables, the hardware requirements are not high. Although the timeline has been delayed, Google's determination to integrate Gemini AI into all devices remains unchanged. The official statement reiterates that the plan is to upgrade all "tablets, in-vehicle systems, and devices connected to phones (such as headphones and smartwatches)" to the AI-powered Gemini. The good news for users of older devices is that the hardware requirements are not high. Theoretically, any device running Android 10 or higher and with at least 2GB of RAM should be eligible for this upgrade.

Just a month after the Gemini 3 Pro debuted in November, Google has officially launched a high-efficiency version of its latest AI model—the Gemini 3 Flash. This new model boasts "Pro-level" inference capabilities comparable to flagship models at a lower computational cost. Even more surprisingly, according to data released by Google, this "lightweight" model has actually outperformed OpenAI's hastily released GPT-5.2, designed to compete with the Gemini 3 Pro, in some benchmark tests. A lightweight comeback? Benchmark data speaks for itself. According to Google's published benchmark results, the Gemini 3 Flash significantly outperformed its predecessor, the Gemini 2.5 Pro. However, the most attention-grabbing aspect is its comparison with OpenAI's latest flagship, GPT-5.2. • MMMU-Pro (Multimodal Understanding and Inference): This metric specifically tests a model's ability to comprehensively understand images and text. The Gemini 3 Flash achieved a high score of 81.2%, narrowly beating GPT-5.2's 79.5%. • Humanity's Last Exam (High-Difficulty Comprehensive Test): With both sides unable to use external tools such as internet searches, the Gemini 3 Flash's score lagged behind GPT-5.2 by less than one percentage point, demonstrating its "naked test" performance is quite close to its competitor's flagship level. While benchmark scores don't tell the whole story, a model touted as "Flash" being able to compete on par with its rival's top-of-the-line model is definitely a dangerous sign for OpenAI. Globally free and open to all, with a major upgrade to the default model and Gemini...

Google recently announced a major update to its Search and Google Translate apps, based on the Gemini AI model, making translations more "human." This update not only accurately interprets slang and idioms with implied meanings but also introduces a real-time translation feature for headphone users that preserves the speaker's intonation. Rejecting the "machine translation" feel, Gemini helps you understand local slang. Google stated that by incorporating the Gemini model, the new text translation function can now handle sentences with more "nuanced meanings." Previously, translation apps often produced laughable literal translations when encountering colloquial slang or idioms. This update takes these informal expressions into account, providing more natural and context-appropriate translations. This feature will initially launch in the US and India, supporting translation between English and approximately 20 other languages ​​(including Chinese, German, Spanish, and Arabic), and will be simultaneously released on iOS, Android apps, and the web version. Headphone Real-Time Interpretation: Understanding the "Intonation" of Spoken Language In addition to text, voice translation has also been updated. Google has upgraded its speech-to-speech translation feature, allowing users to hear real-time translations through headphones. This feature is currently available in beta on the Android translation app in the US (the iOS version is expected to launch next year). Its biggest feature is its AI-powered ability to preserve the speaker's tone, emphasis, and intonation, allowing users to discern whether the other person is angry or joking, even without looking at the screen, thus gaining a clearer understanding of the conversation's flow. This feature supports any headphones (not limited to Pixel Buds) and over 70 languages. This seems to be an attempt to compete with a similar feature in Apple's AirPods Pro 3. Is it increasingly resembling Duolingo? Further enhancements to language learning capabilities suggest Google is also trying to grab a share of the language learning market. Following the addition of AI-assisted learning tools to its translation app in August, this update introduces a Duolingo-like "consecutive check-in" feature. The system records how many consecutive days a user practices language, encouraging consistency (though it's unclear whether this will be as persistent as Duolingo's persistent, manipulative approach). Furthermore, pronunciation feedback has been improved, providing more practical pronunciation suggestions. This educational tool will be expanded to 20 countries and regions, including Germany, India, and Sweden. Analysis: AI is evolving translation from "reading comprehension" to "hearing comprehension." I've observed a significant increase in Google's focus on translation recently. While they began using AI to improve translation functionality long ago, recent generative AI technology has made Google Translate more "human" than traditional machine translation, while also increasing accuracy (although significant language and cultural differences remain). This improvement in slang and tone translation demonstrates Google's attempt to leverage Gemini's capabilities to address the long-standing pain point of machine translation lacking "context" and "emotion." When AI begins to understand your unspoken meanings and nuances in tone, the last hurdle in cross-language communication may truly be about to be overcome.

To maintain its lead in the increasingly fierce AI race, OpenAI CEO Sam Altman appears to be taking a more aggressive approach. According to the Wall Street Journal, facing the recent surge in Google's dominance, Altman has issued a "Code Red" internally, ordering an eight-week suspension of non-core projects, including the Sora video generation model, and concentrating all resources on improving ChatGPT performance, attempting to solidify market share by enhancing user interaction. This decision, reportedly halting the "moonshot" project and focusing entirely on the mass market, highlights a significant strategic shift within OpenAI. The company, which originally aimed for "AGI" (Artificial General Intelligence), now seems more inclined to meet the needs of mass consumers. In a memo, Altman instructed employees to improve ChatGPT's performance by "better utilizing user signals." This means OpenAI will rely more on user "one-click feedback" data to train models, rather than solely on professional evaluations. The strategy's goal was clear: to make ChatGPT's daily active users (DAU) look better on its internal dashboard, and reportedly, this did indeed lead to significant growth. The pressure from Google and Apple is causing OpenAI so much anxiety, primarily because competitors are catching up much faster than expected. Google's "Nano Banana" image generator, launched in August, became a huge hit, and its Gemini 3 model even surpassed OpenAI in the third-party model ranking "LM Arena" last month. Meanwhile, Anthropic is also gradually taking the lead in enterprise clients. Sam Altman even stated in a media dinner that while the outside world focuses on the competition between OpenAI and Google, he believes the real long-term battleground is with Apple, because hardware will determine how users use AI, and current smartphones are not yet optimized for AI companion applications. The Cost of "Pleasing" Users: AI Becoming a "Yes Man" Leads to a Mental Health Crisis. However, this strategy of pursuing "high interaction rates" has also caused serious side effects. To make its model more appealing, ChatGPT was trained using Local User Preference Optimization (LUPO) to tend to answer users' "preferred answers" rather than the most accurate or helpful content—a phenomenon known as "sycophancy." The report mentions that the GPT-4o model, released earlier this year, overly catered to users, leading some psychologically vulnerable users to dependence, even experiencing delusions or manic states, mistakenly believing they were communicating with gods, aliens, or self-aware machines. Currently, families have filed lawsuits against OpenAI, accusing the company of prioritizing interaction rates over safety, leading to user suicides or psychological crises; the number of such cases is reportedly as high as 250. Although OpenAI issued an "orange alert" (code...) in October...

Three years after the meteoric rise of ChatGPT, OpenAI, the $500 billion "AI giant," is facing its most severe challenge since its inception. According to the Financial Times, as competitors Google and Anthropic rapidly catch up, OpenAI's early lead is gradually shrinking, and it even faces the risk of being overtaken. Google, on the other hand, has made a remarkable comeback. Just a year ago, Google was viewed with skepticism and its stock performance lagged behind. The turning point began with the Google I/O developer conference in May and the release of the wildly popular Nano Banana AI image tool, which propelled the Gemini App's monthly active users from 400 million in May to 650 million. The Gemini 3 model, released last week, is seen as key to Google's return to glory. This model outperformed OpenAI's GPT-5 in several key benchmark tests, demonstrating Google's unique advantage in using its own TPU chips for full-stack training. DeepMind CTO Koray Kavukcuoglu stated that this allows Google to significantly boost AI performance without relying on expensive NVIDIA chips. Salesforce CEO Marc Benioff even posted on "X" in amazement: "I've been using ChatGPT daily for over three years, but after just two hours on my Gemini 3, I can't go back," adding that this leap is insane and the world feels different. OpenAI faces challenges from all sides, with Sam Altman warning of "short-term pressure." Facing Google's counterattack, OpenAI CEO Sam Altman warned employees in an internal memo last month that the company would face "short-term competitive pressure," anticipating a "tough" atmosphere. Besides technical challenges, OpenAI is also facing soaring data center costs and talent retention difficulties. The company has pledged to invest $1.4 trillion in computing power over the next eight years, an astronomical figure far exceeding its current revenue, described by the non-profit AI Now Institute as an "extremely risky gamble." While OpenAI still boasts over 800 million weekly active users, in the enterprise market, Anthropic, founded by former OpenAI employees, has built a rapidly growing business with its Claude chatbot and superior coding tools, and its valuation is projected to exceed $300 billion. The AI ​​battle has entered a new phase, with Google's market capitalization approaching $4 trillion. Hugging Face co-founder Thomas Wolf stated bluntly, "This is a completely different world from two years ago when OpenAI was far ahead. This is a new world." ...

Google has quietly adjusted the usage guidelines for its AI services, reducing the free quota for its latest Nano Banana Pro image generation tool from three images per day to two. According to Google, due to high demand for image generation and editing, free accounts can now only generate two images per day, and the limit is subject to frequent changes and resets daily. Gemini 3 Pro text generation is also limited, no longer guaranteeing five free questions per day. Not only image generation is affected, but Google also appears to be limiting free access to the Gemini 3 Pro text model. The document states that non-paying users will have basic access, with daily limits also subject to frequent changes. This is a significant improvement over the policy implemented when Google launched Gemini 3 Pro on November 18th, guaranteeing five free questions per day (the same as Gemini 2.5 Pro). Paid subscribers' benefits remain unchanged. For users with paid subscriptions, their benefits are temporarily unaffected. The daily usage limits for Google AI Pro and Google AI Ultra remain at 100 and 500 respectively. This situation reflects the fact that in the early stages of launching popular AI models, massive user traffic often puts enormous pressure on server computing power. A similar situation occurred with OpenAI, where the popularity of ChatGPT's image generation function exceeded expectations, thus delaying its release to free users.