Latest news with #GoogleDeepMind
NDTV
13 hours ago
- NDTV
Google's AI Chatbot Panics When Playing Video Game Meant For Children
Artificial intelligence (AI) chatbots might be smart, but they still sweat bullets while playing video games that seemingly young kids are able to ace. A new Google DeepMind report has found that its Gemini 2.5 Pro resorts to panic when playing Pokemon, especially when one of the fictional characters is close to death, causing the AI's performance to experience qualitative degradation in the model's reasoning capability. Google highlighted a case study from a Twitch channel named Gemini_Plays_Pokemon, where Joel Zhang, an engineer unaffiliated with the tech company, plays Pokemon Blue using Gemini. During the two playthroughs, the Gemini team at DeepMind observed an interesting phenomenon they describe as 'Agent Panic'. "Over the course of the playthrough, Gemini 2.5 Pro gets into various situations which cause the model to simulate "panic". For example, when the Pokemon in the party's health or power points are low, the model's thoughts repeatedly reiterate the need to heal the party immediately or escape the current dungeon," the report highlighted. "This behavior has occurred in enough separate instances that the members of the Twitch chat have actively noticed when it is occurring," the report says. While AI models are trained on copious amounts of data and do not think or experience emotions like humans, their actions mimic the way in which a person might make poor, hasty decisions when under stress. In the first playthrough, the AI agent took 813 hours to finish the game. After some tweaking by Mr Zhang, the AG agent shaved some hundreds of hours and finished the game in 406.5 hours. While the progress was impressive, the AI agent was still not good at playing Pokémon. It took Gemini hundreds of hours to reason through a game that a child could complete in significantly less time. The chatbot displayed erratic behaviour despite Gemini 2.5 Pro being Google's most intelligent thinking model that exhibits strong reasoning and codebase-level understanding, whilst producing interactive web applications. Social media reacts Reacting to Gemini's panicky nature, social media users said such games could be the benchmark for the real thinking skills of the AI tools. "If you read its thoughts when reasoning it seems to panic just about any time you word something slightly off," said one user, while another added: "LLANXIETY." A third commented: "I'm starting to think the 'Pokemon index' might be one of our best indicators of AGI. Our best AIs still struggling with a child's game is one of the best indicators we have of how far we still have yet to go. And how far we've come." Earlier this month, Apple released a new study, claiming that most reasoning models do not reason at all, albeit they simply memorise patterns really well. However, when questions are altered or the complexity is increased, they collapse altogether.
Economic Times
15 hours ago
- Entertainment
- Economic Times
YouTube Shorts has hit 200 billion daily views: CEO Neal Mohan
Live Events YouTube has crossed a major milestone with its short video feature, YouTube Shorts , now getting around 200 billion views every day, according to CEO Neal Mohan . He shared the update during his keynote address at the 2025 Cannes Lions marks a huge jump from last year, when YouTube reported (March 2024) that Shorts was averaging about 70 billion daily views. Viewership has gone up by nearly 186% in just a also revealed that people are now watching more than 1 billion hours of YouTube on their TVs every day. In May, YouTube topped Nielsen's The Gauge report for the fourth month in a row, making it the most-watched streaming platform in the platform made up 12.5% of all TV viewership, the largest share ever recorded by The Gauge across any streaming service or channel.'For more than half of the top 100 most-watched YouTube channels in the world, TV is their most-watched screen,' Mohan Shorts continues to grow in popularity, YouTube is adding more tools to help creators. Mohan announced that Veo 3, the latest version of Google DeepMind's video generation model, will be launching on the platform later this AI tool helps users create video clips and backgrounds for Shorts. The earlier Dream Screen feature used an older version of Veo, but Veo 3 will bring sharper visuals and the ability to add audio, giving creators more ways to bring their ideas to life.
Deccan Herald
15 hours ago
- Deccan Herald
Google brings AI Mode for voice search on iPhone, Android mobiles
Earlier this year, in March, Google introduced an experimental Artificial Intelligence (AI) Mode in the Search this had a limitation as it only took text requests for AI mode search. Now, the company has added a voice search option called 'Search Live'. Users can ask questions, and it will respond in interaction can be as natural as people talking to each other. However, the generated AI (gen AI) voice will sound a little robotic, as this is a beta feature. The new feature is now available on the Google Search app for iPhone and Android phones in the US. It will be expanded to more regions, including India, in the coming 16: Seven key features you should know about Google's latest mobile how to use 'Search Live' on your phone:Users have to just open the Google app >> tap the new 'Live' icon, and verbally ask their question, like, What are some tips for improving the flowering growth in the rose plant?'The user will hear a helpful AI-generated audio response, and he/she can easily follow up with another question, like, "How many times should I resoil the pot in a year?"The AI chatbot will also offer easy-to-access links right on the phone's screen so the user can dig deeper with content from the AI chatbot also offers transcription of the interaction, and this can be stored in the notepad on the phone for future the coming months, Google plans to introduce even more Live capabilities to AI Mode, including the camera option where users can talk back-and-forth, and users can show the Search app what he/she seeing in real time and retrieve any related information on the a related development, Google DeepMind, the creator of Gemini AI, is reportedly working on a new technology that can mimic the writing of the subscriber and auto reply to people via email, Demis Hassabis, the head of Google DeepMind, confirmed at the SXSW London the next-generation Gmail, users can automate their responses. And, he/she need not have to look at the email, Gmail will intuitively understand the context of the message and respond in an appropriate tone of the owner's writing style. .Gmail replies: Google DeepMind working on AI tool to auto-respond in user's writing the latest news on new launches, gadget reviews, apps, cybersecurity, and more on personal technology only on DH Tech
Time of India
16 hours ago
- Entertainment
- Time of India
YouTube Shorts has hit 200 billion daily views: CEO Neal Mohan
YouTube has crossed a major milestone with its short video feature, YouTube Shorts , now getting around 200 billion views every day, according to CEO Neal Mohan . He shared the update during his keynote address at the 2025 Cannes Lions Festival. This marks a huge jump from last year, when YouTube reported (March 2024) that Shorts was averaging about 70 billion daily views. Viewership has gone up by nearly 186% in just a year. Mohan also revealed that people are now watching more than 1 billion hours of YouTube on their TVs every day. In May, YouTube topped Nielsen's The Gauge report for the fourth month in a row, making it the most-watched streaming platform in the US. The platform made up 12.5% of all TV viewership, the largest share ever recorded by The Gauge across any streaming service or channel. 'For more than half of the top 100 most-watched YouTube channels in the world, TV is their most-watched screen,' Mohan said. Live Events As Shorts continues to grow in popularity, YouTube is adding more tools to help creators. Mohan announced that Veo 3, the latest version of Google DeepMind's video generation model, will be launching on the platform later this summer. Discover the stories of your interest Blockchain 5 Stories Cyber-safety 7 Stories Fintech 9 Stories E-comm 9 Stories ML 8 Stories Edtech 6 Stories This AI tool helps users create video clips and backgrounds for Shorts. The earlier Dream Screen feature used an older version of Veo, but Veo 3 will bring sharper visuals and the ability to add audio, giving creators more ways to bring their ideas to life. Also Read: AI age is here, but human content will remain king: YouTube CEO Neal Mohan
Time of India
a day ago
- Business
- Time of India
Gemini Forged like Achilles: Google Shields Gemini, with Powerful Security Advancements.
What are these threats that AI has grown vulnerable to? Live Events How does Google protect their Gemini models from such threats? Delving a bit deeper into these methods of increasing security: Automated Red Teaming (ART) and Adversarial Fine Tuning: Instruction-Data Separation: Constant Evaluation: As AI expands itself into more industries, the need for AI to be forged like a modern-day Achilles becomes increasingly necessary. For, the more its branches extend into various sectors, the more vulnerable it shall become to the ever-evolving minefield of digital threats. Google DeepMind has unveiled an upgrade to the security safeguards, with a goal to protect its Gemini its newly launched 'White Paper,' Google has laid down its strategic blueprint for combating 'indirect prompt injections' that make agentic AI tools supported by advanced large language models vulnerable to attacks of such kind. Google has made its wants clear to the public, wherein they choose to create AI tools that aren't just capable but also tools that are agents are known for their ability to fulfill straightforward tasks in a moment's time; completion of said tasks, however, involves garnering access to information from various means, such as documents, calendars, or external websites. 'Indirect prompt injection' infects these data sources with 'malicious instructions designed to trick AI into sharing private data or misusing its permissions.'Indirect prompt Injection has become an emerging cybersecurity challenge, AI falls short on identifying the difference between instructions from a genuine user and manipulative commands embedded within the data they prompt injection attacks tend to be complex and often require constant surveillance with the requirement of multiple layers of defense. Rather than combatting these challenges manually, evoking a slow and inefficient result, Google has built an automated system that relentlessly strengthens Gemini's strategy essentially involves the internal team constantly attacking Gemini in relevant ways to pinpoint security weaknesses that Gemini possesses. This technique, in comparison to the others elaborately mentioned in the 'white paper,' helped significantly to increase Gemini's protection rate against 'indirect prompt injection' attacks while the tool is being used by the makes modern-day cyberattacks malicious is their adaptive nature. Upon being struck once by the security safeguards, they return again, with adaptive measures, ensuring they make palpable damage. Hence, basic security measures work well against non-adaptive cyberattacks, the kind that stands in contrast to the one explained to combat complex attacks such as this, Gemini's security enhancements ensure focus on both proactive and reactive strategies:'ART generates effective indirect prompt injections targeting sensitive information.' This essentially mimics the tactics of real-world adversaries, which in turn teaches Gemini to ignore such malicious embedded instructions and follow the original user request, resulting in the model only providing the correct and secure answer it is meant to provide. Further, this form of training lets the model innately understand how it must handle compromised information that evolves over time as part of adaptive safeguard helps Gemini differentiate between a command fed to the model by the genuine user and prompts that are embedded with malicious instructions. This is deemed to be an essential defense line against prompt the attacks' adaptive capabilities, it asks for constant surveillance; therefore, the system is tested using a dynamic feedback loop of continuous understands that this isn't a 'solved' problem but merely a step forward into the minefield. As generative AI becomes a pivot to search, productivity tools, assistance, and more, the stakes for secure and trustworthy AI are higher than ever. Therefore, Gemini's upgrade marks a key milestone in this AI race, where they ensure that powerful tools such as these remain loyal to their wielder.
