Startups & Ventures

⚡️ Google launches Gemini 3 Flash Google has introduced Gemini 3 Flash, positioning it as a frontier-level model optimized for speed, cost, and low latency, aimed at real production workloads rather than showcase demos. 🔸 Google claims fast reasoning and strong coding performance, with full multimodality designed for day-to-day deployment, not experimental use. 🔸 On GPQA Diamond and Humanity’s Last Exam, benchmarks targeting PhD-level reasoning, Gemini 3 Flash performs confidently, signaling parity with higher-end models on hard academic tasks. 🔸 On MMMU Pro, a challenging multimodal benchmark, Flash nearly matches Gemini 3 Pro, despite being positioned as a lighter, faster model. 🔸 In efficiency terms, Flash is significantly faster than Gemini 2.5 Pro, while also consuming fewer tokens, improving both latency and cost. 🔸 On SWE-bench Verified, Gemini 3 Flash outperforms not only the entire 2.5 series but also Gemini 3 Pro, highlighting its strength in real-world software engineering tasks.

Gemini 3 Flash is about making frontier-level intelligence cheap, fast, and deployable, which is where the real competition is shifting.

📊 Powered by Crypto Insider

🧬 OpenAI put GPT-5 into a real biology lab and it made DNA cloning 79× more efficient OpenAI has run a first-of-its-kind experiment with GPT-5 operating inside a real wet lab, working in a closed loop with scientists at Red Queen Bio to optimize a core molecular biology process. 🔸 This wasn’t a simulation. GPT-5 proposed hypotheses and protocols, humans and lab robots executed them, measured real biological outcomes, and fed the results back to the model for the next iteration. 🔸 The task was Gibson Assembly, a standard DNA cloning technique taught in basic molecular biology. The optimization metric was simple and objective: the number of successful colonies. 🔸 Gibson Assembly is old, well-studied, and rarely optimized in practice, doing so is labor-intensive and usually yields only 2–3× improvements, if any. 🔸 Over multiple experimental cycles, GPT-5 achieved a 79× increase in efficiency, with results that were stable and reproducible across runs. 🔸 The key insight: GPT-5 suggested adding two known proteins, RecA and gp32 into the same reaction, a combination no one had previously tested specifically for cloning, despite both proteins being well understood individually. 🔸 This isn’t a scientific revolution or a new biological principle, more like the output of a very strong PhD student focused on a narrow, technical problem. 🔸 What is new is the role shift: GPT-5 wasn’t just generating text or running agents, it became an active participant in a physical scientific process, learning from the real world in a feedback loop.

The takeaway isn’t “AI solved biology”, it’s that language models are starting to cross the boundary between software and reality. And once that loop is open, the pace of experimentation changes dramatically.

📊 Powered by Crypto Insider

✅ OpenAI responds to Nano Banana with GPT Images OpenAI has unveiled GPT Images, its new image generation and editing model, positioning it as a direct answer to fast-moving competitors like Nano Banana. 🔸 OpenAI promises photorealistic output and much stronger instruction following, especially for complex prompts where small details actually matter. 🔸 A major focus is high-quality image editing including multi-step edits where faces, identities, or specific elements must remain unchanged across revisions. 🔸 Image generation is now up to 4× faster, signaling a push toward real-time creative and agentic workflows rather than slow, batch-style generation. 🔸 In ChatGPT Images, the model is already available to all users, including free users (usage limits exist, but haven’t been clearly disclosed yet). 🔸 For developers, OpenAI has released GPT Image 1.5 in the API, positioning it as a production-ready image model rather than a demo toy.

OpenAI is clearly treating image generation as core infrastructure, not a side feature.

📊 Powered by Crypto Insider

Robots are not yet ready to completely replace humans, even in warehouses. 📊 Powered by Crypto Insider

Robots are not yet ready to completely replace humans, even in warehouses. 📊 Powered by Crypto Insider

🔐 A popular VPN extension was secretly collecting AI chats A security investigation by Koi Security revealed that the Urban VPN browser extension has been silently collecting users’ conversations with AI tools, raising serious privacy concerns around browser add-ons marketed as “free” and “secure.” 🔸 Millions of users may be affected. Urban VPN has over 6 million installs on Chrome, and together with related extensions from the same publisher, the total reach exceeds 8 million users across Chrome and Edge. 🔸 AI conversations were fully captured. The extension intercepted entire AI chats, including user prompts, AI responses, timestamps, and metadata. Affected platforms reportedly include ChatGPT, Claude, Gemini, Copilot, Perplexity, Grok, and Meta AI. 🔸 The data capture was deeply embedded. Researchers found that the extension injected hidden scripts into AI websites, overriding core browser functions like fetch() and XMLHttpRequest() allowing it to read conversations before they appeared on screen. 🔸 “AI protection” claims were misleading. While Urban VPN promoted an “AI protection” feature, the investigation found that data harvesting ran continuously in the background and could not be disabled without uninstalling the extension entirely. 🔸 Updates rolled out silently. The behavior was introduced in a July 2025 update and automatically deployed to users, meaning many had no idea their AI conversations were being collected. 🔸 The data was monetized. According to the report, captured conversations were shared with a data broker affiliate and used for analytics and commercial purposes.

Free VPNs and browser extensions can come with hidden costs and in the AI era, your private conversations may be far more exposed than you think.

📊 Powered by Crypto Insider

✅ Google launches Gemini Agent, its first true AI assistant that acts for you Google has rolled out Gemini Agent, a new AI system designed to operate on behalf of users directly in the browser, marking a major step toward practical, action-oriented AI assistants. 🔸 Premium and limited launch. For now, Gemini Agent is available only in the U.S. and exclusively to users on the Gemini Ultra plan, priced at $250 per month, positioning it as a high-end, early-access product. 🔸 An agent that actually does things. Unlike chat-based assistants, Gemini Agent can navigate websites, gather information, compare prices, and complete tasks such as booking hotels, rental cars, and other services, effectively acting as a digital concierge. 🔸 Built on Gemini 3. The agent is powered by Gemini 3, Google’s latest model, enabling it to understand complex instructions, reason across multiple steps, and interact with real web interfaces rather than just generating text. 🔸 Cross-platform support from day one. Gemini Agent works on desktop browsers and mobile apps, allowing users to hand off tasks seamlessly across devices.

Google is moving beyond AI that answers questions toward AI that executes workflows, a direct challenge to startups building autonomous agents and a signal that consumer AI is entering its “do the work for you” phase.

📊 Powered by Crypto Insider

🔥 Red Bull made the biggest Tetris in the world The finalists of the Tetris tournament played on 2,800 drones, which displayed falling figures in real time 📊 Powered by Crypto Insider

🧠 “Google’s biggest mistake was underestimating transformers” — Sergey Brin Stanford has published a revealing Q&A with Sergey Brin, where the Google co-founder reflects on how the company that invented key AI breakthroughs still lost momentum to OpenAI. 🔸 Brin argues that OpenAI’s success wasn’t purely technical. The real edge was belief, belief in scaling models aggressively and in making chatbots a core product, not just a research demo. 🔸 Google, he says, didn’t fully commit to transformers after publishing the landmark 2017 paper “Attention Is All You Need.” The architecture existed, but the company wasn’t willing to invest billions early enough to see where scaling could lead. 🔸 The same hesitation applied to chatbots. Google avoided releasing them widely because they “sometimes talked nonsense,” prioritizing product safety and brand risk over speed and experimentation. 🔸 OpenAI took the opposite approach: ship early, iterate in public, and accept imperfections, a strategy that ultimately captured users, mindshare, and capital. 🔸 Brin’s most striking thought experiment: what if Google had kept transformers closed and poured a few billion dollars into scaling them back in 2017?

The lesson is uncomfortable but clear: having the breakthrough matters less than having the conviction to bet on it early — and publicly.

📊 Powered by Crypto Insider

⚛️ Google is building nuclear data centers and AI is the reason Artificial intelligence needs power, enormous amounts of it and Big Tech is increasingly turning to nuclear energy as the only realistic long-term solution. Google is now making that bet explicit. 🔸 Google announced a major expansion of its partnership with NextEra Energy, one of North America’s largest power producers, to build gigawatt-scale data center campuses across the US. The first three sites are already under active development. 🔸 The partnership isn’t new. About 3.5 GW of generation capacity is already operating or under contract, and the companies recently agreed to add 600 megawatts of clean energy to the Oklahoma grid to support Google’s growing infrastructure. 🔸 The most notable shift is the planned restart of the Duane Arnold Energy Center nuclear plant in Iowa. NextEra has filed with FERC to regain grid connection rights, a sharp reversal from earlier plans to replace the site with solar generation. 🔸 Beyond power generation, Google Cloud will support NextEra’s digital transformation, deploying AI models like TimesFM 2.5 for time-series forecasting and WeatherNext 2 for weather prediction to optimize grid operations and reduce failure risks. 🔸 The first commercial product from this collaboration is expected to launch on the Google Cloud Marketplace by mid-2026, signaling that energy optimization itself is becoming a cloud software business.

Renewables alone can’t keep up with AI, and nuclear is quietly re-emerging as the backbone of next-generation data centers.

📊 Powered by Crypto Insider

📱An iPhone case with a screen on the back and extra storage Accessory maker Dockcase has unveiled Selfix, a case for the iPhone 17 Pro that tackles two common pain points at once: better selfies and more storage. 🔸 On the back of the case is a 1.6-inch AMOLED display (480×480), fully touch-enabled, that works as a rear viewfinder. Flip the phone around, frame yourself on the screen, and shoot selfies using the main camera system instead of the front camera. 🔸 The result is significantly higher-quality selfies, leveraging Apple’s primary sensors rather than the lower-end front-facing camera. 🔸 Selfix draws power directly from the iPhone, meaning slightly higher battery drain, but no extra charging, pairing, or setup is required, snap it on and it works. The case also doubles as a pop socket for grip. 🔸 The second feature is a microSD card slot supporting up to 2TB, offering expandable storage, something iPhones still lack natively. 🔸 Rear displays aren’t new, Android phones like the LG V10, Meizu Pro 7, and more recently Xiaomi 17 Pro have tried similar ideas but Selfix is unusual because it delivers the concept as a simple accessory, not a custom phone design.

Selfix will launch in Oat White, Blush Pink, and Midnight Black. Pricing and release timing haven’t been announced yet, but the case hints at a future where accessories meaningfully extend what phones can do without replacing them.

📊 Powered by Crypto Insider

🌐 A browser that builds apps instead of showing links Instead of returning a list of blue links, Disco turns your search query into a working mini-app, generating tools, maps, and calculators on the fly based on what you’re trying to do. 🔸 Ask Disco to plan a trip to Japan, and it creates an interactive route map with attractions. Say you’re moving cities, and it assembles a weight calculator and carrier price comparison tailored to your move. 🔸 Disco is an experimental browser from the Chrome team, born out of an internal Google hackathon and now rolling out publicly through Chrome Labs. 🔸 The browser is built around GenTabs, a concept where the Gemini AI interprets your request, opens relevant websites in the background, and stitches their information into a personalized, task-specific app. 🔸 The system works both ways: if you open a hotel or article in a neighboring tab, GenTab pulls that data in automatically and updates your plan. 🔸 Early versions struggled because users ignored links shown inside chat and kept talking to the bot. The team fixed this by having the browser itself open sources, nudging users to actually explore the web and early data suggests it worked. 🔸 Big questions remain unanswered, including whether GenTabs can be saved, shared, or exported to tools like Google Docs. Even the team admits these decisions haven’t been made yet.

Disco isn’t meant to replace Chrome, it’s a standalone experiment testing whether the future of the web lies in temporary, AI-assembled apps built for a single task, then discarded.

📊 Powered by Crypto Insider

👓 China’s Li Auto just unveiled smart glasses that control your car Chinese EV maker Li Auto (Lixiang) has introduced its own pair of smart glasses designed to act as a remote control and AI interface for its vehicles, expanding the company’s in-car ecosystem beyond the dashboard. 🔸 Through Li Auto’s Lixiang Livis system, users can summon their car from a parking lot, open the trunk, pre-cool or heat the cabin, and manage other vehicle functions directly from the glasses. 🔸 The glasses also double as a first-person camera, with Sony image sensors built into the frame, allowing drivers to record hands-free POV video. 🔸 An AI assistant is integrated into the device, enabling voice commands and contextual assistance while on the move. 🔸 Li Auto claims the glasses offer up to 19 hours of battery life, positioning them as an all-day wearable rather than a short-use gadget.

The launch underscores how Chinese EV makers are increasingly blending cars, AI, and wearables into a single platform, turning vehicles into extensions of personal computing rather than standalone machines.

📊 Powered by Crypto Insider

🔒Google Translate will add real-time translation for wireless headphones Google is preparing a major update to Google Translate that will enable real-time speech translation through wireless headphones, bringing hands-free, conversational translation to everyday use. 🔸 The feature will work similarly to Apple’s live translation tools, instantly playing translated speech in the user’s headphones as another person speaks a different language. 🔸 Powering the upgrade is Google’s Gemini model, which can understand and translate up to 70 languages, including Russian. 🔸 Unlike some competing solutions, Google’s implementation will not require dedicated or branded hardware. The company says the feature will work with any wireless headphones. 🔸 Google plans to roll out the real-time headphone translation feature in 2026.

The move positions Google Translate for more natural, real-world conversations and could significantly expand how people use translation tools in travel, work, and daily communication.

📊 Powered by Crypto Insider

🚀 With iOS 26.2, Apple adds new control to roll back Liquid Glass on the Lock Screen Apple has begun rolling out iOS 26.2, a point update that gives iPhone users more control over the company’s controversial Liquid Glass design and introduces a set of functional improvements across the system. 🔸 The headline change in iOS 26.2 lets users adjust the transparency of the Lock Screen clock introduced with Liquid Glass, the translucent interface style Apple rolled out earlier this year. This new slider is part of Apple’s continued response to user feedback that some Liquid Glass effects made elements harder to read. 🔸 Beyond design tweaks, iOS 26.2 includes a variety of feature updates. Users can now create one-time AirDrop codes to share content with people not in their contacts, add alarms in the Reminders app, and view Apple Music lyrics offline. The update also brings enhancements to Podcasts, navigation refinements in Apple News, and updated sleep metrics for Apple Watch users. 🔸 The broader iOS 26 series has included iterative adjustments to Liquid Glass since its introduction, including previous opacity sliders and tinted options, as Apple balances its new design language with readability and usability concerns.

This latest release aims to refine the user experience with practical controls and feature improvements that respond to user feedback while maintaining Apple’s design direction.

📊 Powered by Crypto Insider

🚀 ChatGPT-5.2 is now available for free and here’s what’s included The latest and most powerful version of ChatGPT, GPT-5.2, has already been introduced on LMArena, giving users open access without the usual paywalls or limits. 🔸 Completely free access: All core features are available with no usage restrictions. 🔸 Built-in Canvas: Users can generate websites, full applications, and even games directly inside an interactive workspace. 🔸 No VPN required: The model is accessible out of the box, with no regional workarounds or technical tricks needed.

If this rollout holds, LMArena has effectively turned GPT-5.2 into one of the most accessible high-end AI tools currently available, lowering the barrier for developers, creators, and experimenters worldwide.

📊 Powered by Crypto Insider

✈️ Riyadh Air wants to become the world’s first fully AI-native airline built from day one Saudi Arabia’s new national carrier, Riyadh Air, says it has been designed entirely around artificial intelligence, positioning itself as one of the most technologically ambitious airline launches in decades. 🔸 The airline is being built AI-first from scratch, rather than retrofitting legacy systems. Every employee will operate inside a personalized digital workplace centered on an AI chat interface that supports daily operations and decision-making. 🔸 For flight crews, AI systems will deliver real-time insights into passenger preferences, enabling more tailored onboard service. Voice control, automated planning, budgeting, and live demand forecasting are already embedded across operations. 🔸 Riyadh Air says initial flights have already begun, with full commercial service planned for early 2026. By 2030, the airline aims to connect 100+ destinations worldwide, aligning with Saudi Arabia’s Vision 2030 aviation push. 🔸 The project took three years to develop with IBM Consulting and integrates 59 core workflows across more than 60 technology partners, including Adobe, Apple, FLYR, and Microsoft.

If it executes as planned, Riyadh Air won’t just be a new airline, it could become a real-world test of whether AI-native operations can outperform traditional aviation models at global scale.

📊 Powered by Crypto Insider

📚 A massive set of 500 curated prompts from Nano Banana Pro is now available A new collection of 500 hand-picked prompts has dropped, a large, structured table covering ideas, workflows, and shortcuts for nearly every creative or technical task. 🔸 The library spans writing, coding, research, design, productivity, and niche use cases, making it a quick reference for anyone looking to boost prompt quality or spark new ideas. 🔸 Each entry includes a link to the original prompt author’s profile, offering an easy way to explore more of their work and discover additional templates, tools, and inspiration. 🔸 The table is designed for fast scanning: clear categories, examples, and ready-to-copy formulations that work across different models and tools.

For anyone building with AI daily, this looks like one of the strongest prompt collections released so far, both as a resource hub and a gateway to new creators.

📊 Powered by Crypto Insider

🥸 What your browser secretly leaks and the tools to uncover it A handful of inspection tools can show exactly what information your browser exposes every time you load a webpage from fingerprints to tracking vectors you didn’t know existed. 🔸 Browserleaks gives a full snapshot of everything your browser reveals by default, from IP metadata to canvas fingerprints. 🔸 CreepJS evaluates how uniquely identifiable your setup is, showing how easy it is to track you across sessions. 🔸 Cover Your Tracks tests how well your browser resists fingerprinting and whether privacy protections are actually working. 🔸 FingerprintJS demonstrates how websites identify visitors and highlights the specific attributes your browser leaks. 🔸 WebBrowserTools offers a suite of lightweight tests for checking privacy, fingerprint resistance, and general security posture. Use these tools and you’ll know exactly what slips through, before anyone else does. 📊 Powered by Crypto Insider

⚡️ OpenAI releases GPT-5.2 with major gains in reasoning and coding OpenAI has launched GPT-5.2, an upgraded model that delivers substantial improvements in coding, mathematics, long-context reasoning, and image understanding. 🔸 On benchmarks, GPT-5.2 hits 100% on AIME 2025 (up from 94% in GPT-5.1) and reaches 80% on SWE-bench Verified (versus 76.3% for 5.1), marking some of the model’s strongest performance jumps to date. 🔸 OpenAI continues to compare its new model only to earlier internal versions not rivals but based on public metrics, GPT-5.2 appears to edge ahead of Google’s Gemini 3 Pro in several categories. 🔸 In ChatGPT, GPT-5.2 Instant, Thinking, and Pro begin rolling out today, with the API already available. Pricing has increased slightly from GPT-5.1 and GPT-5.

GPT-5.2 signals a real competitive push from OpenAI, the kind of leap that only intense pressure from peers can produce.

📊 Powered by Crypto Insider