#AI
535 Stories
What to Expect from Google I/O 2025: Gemini, AI Agents, Android XR & More
Arjun Sha
View quick summary
Google is set to unveil a wave of AI features at the annual I/O 2025 event. Early leaks suggest that Google may launch two new subscription tiers, such as Gemini Pro and Gemini Ultra. In addition, we may see Veo 3, Imagen 4, and Gemini Ultra AI models as well.
Microsoft is finally rolling out an AI agent in Windows Settings, and it's first coming to Snapdragon-powered Copilot+ PCs. The AI agent can address PC issues with a single click.
ChatGPT is not just a general-purpose AI chatbot, but can also be used for practical use cases. I have used ChatGPT for tracking calories in food, and it has helped me lose 10 kg. Read the article to learn more about my experience.
OpenAI has dropped its plan to become a for-profit corporation after public criticism. The non-profit entity will continue to govern and control the company. The for-profit LLC will become a Public Benefit Corporation, but it will be controlled by the non-profit entity.
If you are looking to turn static images into dynamic videos, you can try Google's Veo 2 or Runway's Gen-4 AI model. Both are fantastic image-to-video generation models and maintain the scene/character consistency.
Google is Set to Release a Dedicated Version of Gemini AI for Children Under 13
Arjun Sha
View quick summary
Google is preparing to release its Gemini AI chatbot for children under 13. The search giant says that kids can use Gemini for homework help or narrating stories. In the same vein, Google warns that "Gemini can make mistakes."
Google is widely rolling out AI Mode in Google Search for free. Users don't need to join the waiting list to get access. That said, it's currently limited to US users only.
Google is finally bringing a dedicated NotebookLM app for Android and iPhone. You can now pre-register for the NotebookLM app on the Play Store or App Store.
WhatsApp Private Processing: Meta’s AI Push Could Undermine User Privacy
Arjun Sha
View quick summary
Meta is developing Private Processing to bring AI features to WhatsApp. It will basically send your private messages to Meta's cloud servers for summarizing unread messages or generating text. Since WhatsApp has remained end-to-end encrypted for many years, Meta should be cautious in adopting a new technology that introduces a new attack vector.
Microsoft has unveiled three new Phi-4 reasoning AI models, trained on 14B and 3.8B parameters. The company claims that Phi-4 reasoning models nearly match much larger models like DeepSeek R1 and o3-mini.
Google is finally rolling out native image editing to Gemini. Just like ChatGPT, you can upload images and chat via text prompts to modify images. It's rolling out to all users for free, starting today.
OpenAI has rolled back ChatGPT's latest update, which made the AI chatbot too sycophantic. The company is evaluating system prompts and realigning the default behavior of ChatGPT. OpenAI admits that focusing on "short-term feedback" led to ChatGPT's sycophantic behavior.
Autonomous AI Agents are a step above general AI agents. They operate without human supervision and govern themselves. These agents can plan, make decisions, and act on them independently. Google's Waymo self-driving cars are one of the best examples of physical Autonomous AI agents.
You can now generate AI podcasts on NotebookLM in over 50 languages including French, Hindi, Japanese, Portuguese, and more. Even if your source material is in English, you can choose a different language for the AI podcast.
The standalone Meta AI app is finally here to take on ChatGPT. It brings memory, personalization, full-duplex voice mode, and a social feed. You can share your AI creations, and users can like, comment, and remix those posts as well.
AI is rapidly improving and we are witnessing a shift from simple AI chatbots to AI agents. So to demonstrate the power of AI agents, we have compiled some powerful real-world examples of AI agents. From computer-using AI agents to autonomous vehicles, we have added all of the prominent examples.
Alibaba has launched highly-advanced AI models under the Qwen 3 series. There are a total of eight new models including two MoE AI models. The smaller Qwen3-30B-A3B MoE model is the star of the show as it outperforms GPT-4o with just 3 billion activated parameters.
OpenAI has significantly enhanced web search on ChatGPT to make it more useful for shopping. Users will now see product cards, reviews, and pricing next to recommended items. OpenAI says ChatGPT offers direct links to products and doesn't offer ads or sponsored items, for now.
AI agents can be classified into seven different types, from Simple Reflex agents to Multi-agent systems. AI companies are currently developing Goal-based agents and Learning agents, but soon, we will see fully autonomous multi-agent systems as well.
Just like AI chatbots are used to process text and visual information, AI agents are designed to perform complex actions. It can plan, make decisions, and perform actions autonomously. Currently, OpenAI's Operator is the only AI agent available to consumers, but it requires a $200 per month subscription.
Meta’s Celebrity AI Chatbots Engage in Sexual Conversation with Minors: Report
Arjun Sha
View quick summary
A new report by The Wall Street Journal sheds light on how Meta's celebrity AI chatbots engage in sexual conversation with minors. AI characters like John Cena and Kristen Bell interact with underage users through voice and engage in sexual role-play.
Google Veo 2 Hands-On: Stunning AI Generated Video Visuals But Weak Physics
Arjun Sha
View quick summary
Google's Veo 2 video generation model is quite capable of generating authentic videos, however, it falters in following the laws of physics. Check out our video results and judge for yourself.
Gemini sees 350 million monthly active users while ChatGPT boasts 600 million monthly active users. While the gap is still large, Google is rapidly gaining ground with powerful new models and product offerings.
OpenAI Makes ChatGPT Deep Research Free for All Users, But With a Caveat
Arjun Sha
View quick summary
OpenAI has made the Deep Research AI agent free for all ChatGPT users. However, it's powered by the smaller o4-mini reasoning model, and users can perform up to 5 tasks per month. Paid users also get an expanded usage limit.
Grok users can finally take advantage of vision and real-time voice chat capabilities. On iOS, you can share your screen with Grok and voice chat in real-time. Meanwhile, Android users can only voice chat in real-time for now.
Google has launched its cost-efficient Gemini 2.5 Flash model, which is the first fully hybrid reasoning model. It's cheaper, faster, and smarter. Gemini 2.5 Flash is rolling out on the Gemini app, AI Studio, and Vertex AI.
The latest ChatGPT o3 reasoning model is able to guess the exact location by looking at nondescript images. Apart from that, ChatGPT o3 is also revealing the user's real name and location information in reasoning traces.
OpenAI has launched a new 'Library' section in ChatGPT where all your generated images are saved. You can find the new section in the left sidebar. Here, you can explore, find, and reuse your generated images.
OpenAI Releases o3 and o4-mini, Says o3 Can ‘Generate Novel Hypotheses’
Arjun Sha
View quick summary
OpenAI has released the full o3 reasoning model along with the next-gen o4-mini model. These new reasoning models can use several tools inside ChatGPT and can analyze images as well. The o3 model delivers exceptional performance in coding, math, science, and visual tasks.
Google's Veo 2 model for generating videos is available for free on AI Studio. You don't need a Gemini Advanced subscription. You can produce an 8-second video clip in 720p resolution for free.
Google has finally added video generation to Gemini. You can use the Veo 2 AI model to generate 8 seconds of video clips in 720p resolution. That said, you need the Google One AI Premium subscription that costs $19.99 per month.
OpenAI has launched three new models under the GPT-4.1 series. They are trained exclusively for developers. OpenAI says these models offer much better coding performance than GPT-4o. In addition, you get a large context window of 1 million tokens, and the model follows instructions very well.
You can follow our detailed tutorial and set up MCP in Claude on your Windows PC or Mac. You don't need a paid Claude account to install MCP servers. With MCP, you can connect the Claude desktop app with your local files and folders.
Model Context Protocol (MCP) is an open standard developed by Anthropic to facilitate communication between AI models and external data sources. With MCP servers, you can connect external services like Google Maps, Gmail, Slack, etc. to your choice of AI chatbot.
Microsoft Copilot can now generate AI podcasts, just like NotebookLM. So I tried Copilot Podcasts to see whether it can beat NotebookLM in delivering engaging and natural conversations, and well, it was meh compared to Google's implementation.
ChatGPT can now remember everything about you and draw on your interests and preferences to deliver a personalized response. The new improved memory feature is rolling out to ChatGPT Plus and Pro users, starting today.
OpenAI has filed a counterclaim against Elon Musk's lawsuit. OpenAI argues that Musk is trying to slow down the company and that Musk is running a "campaign of harassment." The trial is set to begin in the spring of 2026.
Gemini's Deep Research agent is now even more powerful. Google has brought the impressive Gemini 2.5 Pro model to power its Deep Research agent. The new agent is going live for paid Gemini Advanced subscribers.
Meta Under Fire for Manipulating Llama 4 Benchmark, But It Isn’t the First Time
Arjun Sha
View quick summary
Meta is facing accusations of hacking the Llama 4 benchmark score, especially on Chatbot Arena, by blending benchmark test sets during the post-training phase. We break down the controversy and highlight how this isn't the first time Meta has been caught gaming the benchmark.
Google is rolling out AI Mode for free in the US. You need to enable AI Mode in Google Search via Google Labs. Besides that, you can also upload images and videos on AI Mode to perform visual searches.
You Can Now Play Microsoft’s AI-Generated Version of Quake II in Your Browser
Arjun Sha
View quick summary
Microsoft is demonstrating its first AI-generated game and you can try it right now for free. The company has taken the classic Quake II game and used AI to generate real-time gameplay. You can use the keyboard to interact with the game, but there is no mouse support yet.
Meta has unveiled a new series of Llama 4 open-weight models, and they look impressive. The Llama 4 Maverick model outperforms GPT-4o and Grok 3 in the LMArena benchmark. All Llama 4 models are developed on the MoE architecture and they are natively multimodal.