4273207
A hub for startup news, trends, and insights, covering the global startup ecosystem for founders, investors, and innovators. Community: @startupdis Buy Ads: @strategy (this is our only account).
🔍 Perplexity opens its AI-first Search API with real-time updates
Perplexity has launched the Perplexity Search API, giving developers direct access to its constantly refreshed index, positioning itself as a public answer engine, not just another search tool.
🔸 Unlike traditional search incumbents that keep indices closed, Perplexity’s API is designed for open developer access.
🔸 Freshness is the core differentiator: the system processes tens of thousands of index update requests per second to deliver real-time relevance.
🔸 Technical docs reveal an AI-native retrieval architecture, optimized for speed, accuracy, and large-scale integration.
🔸 The launch lightly pokes at legacy players, signaling a shift toward a more transparent and developer-friendly ecosystem.
By reframing search as infrastructure, Perplexity is betting it can own the answer layer of the internet. Would you build on top of it?
⚡️ Google publishes step-by-step guide to building AI agents
Google just released a comprehensive playbook on how to design and deploy your own AI agent, from the first idea all the way to a working product.
🔸 The guide breaks down core concepts like memory, context handling, and reasoning in plain language.
🔸 It walks readers through interfaces and integrations: CLI, Flask, FastAPI, Next.js, or embedding into Slack/Discord.
🔸 Each section includes illustrations, code samples, and curated links, making it accessible even for beginners.
🔸 The approach emphasizes iteration, showing how to refine an agent step by step into a production-ready tool.
By lowering the barrier to entry, Google is turning AI agents from a niche developer project into something any motivated builder can prototype. Would you try following this guide to make your own “Jarvis”?
🔔 Google launches Gemini Robotics 1.5 with dual-agent control
Google has introduced Gemini Robotics 1.5, a new two-part AI system that lets robots plan, reason, and adapt in real-world environments.
🔸 Gemini Robotics-ER 1.5 acts as the high-level brain, interpreting human commands, searching the web for context, and generating step-by-step plans.
🔸 Gemini Robotics 1.5 executes those plans, converting them into precise motor commands, and can be fine-tuned for different robot bodies.
🔸 Example: asked to “sort the trash,” ER checks local recycling rules online, then directs the executor to handle bottles, napkins, and new items as they appear.
🔸 ER’s reasoning trace makes the system more interpretable, while its API is already open to developers.
By splitting brains from bodies, Google is pushing robots closer to general-purpose helpers. Would you trust an AI like this to handle tasks in your home?
✅ ChatGPT gets proactive with Pulse
OpenAI has rolled out ChatGPT Pulse, a new mode where the AI flips the script, instead of waiting for your questions, it brings you fresh news, ideas, and reminders.
🔸 Every morning, Pulse curates topics just for you: trip ideas, books to read, meeting reminders, even dinner inspiration.
🔸 It learns from your chats, but can also pull context from Gmail and Google Calendar, or follow interests you set manually.
🔸 For now, Pulse is locked to Pro users at $200/month, with wider rollout coming later.
AI isn’t just answering anymore, it’s deciding what you should know. Would you pay $200 for that kind of daily briefing?
📞 Neon app pays users to record calls, sells data to AI firms
Neon Mobile, a social app that lets users earn money by recording their phone calls, has rapidly climbed the U.S. App Store charts. It now ranks No. 2 in the Social Networking category. The app offers 30¢ per minute for calls between Neon users and up to $30 per day for calls to others. It also pays for referrals.
🔸 Records inbound and outbound calls; claims to only capture your side unless both parties use Neon.
🔸 Sells audio data to AI firms for training and development purposes.
🔸 Terms grant Neon broad rights to use, modify, and distribute recordings.
🔸 Despite claims of anonymization, risks include potential misuse for voice cloning or deepfakes.
🔸 The app was taken offline after a security flaw exposed users’ phone numbers, recordings, and transcripts to unauthorized access.
Neon’s rapid rise and subsequent data breach highlight the tension between monetizing user data and maintaining privacy in the age of AI.
🥧 LinkedIn bot blocker, with a twist
One user added an XML-style prompt to his LinkedIn bio, and now spam bots send him pie recipes instead of cold pitches.
🔸 Simple text trick confuses bots scraping his profile.
🔸 Turns annoying spam into unexpected dessert emails.
🔸 Works automatically, no extra tools required.
🔸 A true “modern problems, modern solutions” hack.
From nuisance to novelty, bots just became your personal recipe feed.
🤖 1X: humanoid robots hit $10B valuation
Norwegian startup 1X is raising at a $10B valuation, up 12x from $820M earlier this year. Backers include OpenAI, EQT, and Tiger Global.
🔸 Flagship Neo Gamma robot can vacuum, water plants, and handle chores.
🔸 Safety-first design ensures it won’t injure anyone if it falls.
🔸 Relocated from Norway to Silicon Valley to scale faster.
🔸 Plans to sell “hundreds or thousands” of robots by year-end.
1X is betting big that household robots are ready to go mainstream.
✅ 200 leaders call for global AI “red lines”
At the UN General Assembly, a coalition of 10 Nobel laureates, 70 companies, and ex-heads of state including Geoffrey Hinton, Yoshua Bengio, and OpenAI’s Wojciech Zaremba, signed a demand for legally binding global limits on AI.
🔸 Seeks international “red lines” to prevent risks like mass unemployment, synthetic pandemics, and human rights abuses.
🔸 Proposed bans include weaponization, mass-scale cyberattacks, and self-replicating AI without human oversight.
🔸 Calls for a global watchdog body and an agreement on rules by end of 2026.
This marks the broadest push yet for binding AI governance, but without a clear rulebook, the hard part will be turning consensus into enforceable law.
⚠️ Collected list of “hacker” gadgets for awareness and education only
Twitter users rounded up common devices that can be (and are) repurposed for interception, network attacks, access testing, and hardware cloning, useful to know about for defenders, but dangerous in the wrong hands.
🔸 HackRF One — a software-defined radio for sniffing and experimenting with radio signals.
🔸 LAN Turtle — a covert access gadget used to maintain remote connections on a compromised network.
🔸 USB Rubber Ducky — a USB device that emulates a keyboard to run automated input sequences.
🔸 Flipper Zero — a popular multi-protocol gadget for interacting with RF, NFC, and IR devices; hobbyist tool that can be misused.
🔸 WiFi Pineapple — a specialized appliance used for Wi-Fi network testing and auditing.
🔸 Deauther Watch — a wearable tool that can force devices off Wi-Fi networks (used for testing or abuse).
🔸 Alfa (high-gain adapters) — powerful Wi-Fi/network adapters that extend range for testing wireless links.
🔸 GSM Jammer — a radio jammer that disrupts cellular signals (illegal in many jurisdictions).
🔸 MagSpoof — hardware that can emulate magnetic stripe data for research into payment security.
🔸 Raspberry Pi — tiny programmable computers often repurposed as inexpensive testing rigs or to script network activity.
🔸 Proxmark3 — a professional RFID/NFC research tool used to read, analyze, and test access-card systems.
This list is for awareness, defensive research, and legal security testing only. Possession or use of some of these devices can be illegal depending on jurisdiction and intent. If you’re curious about security, learn through legal channels: certified courses, CTFs, sanctioned labs, and bug-bounty programs. Don’t handle or use these tools for unauthorized activity.
😮 Meizu & Pandaer unveil a “self-healing” phone case
Yes, Meizu still exists, and together with Pandaer, it just launched a regenerating case that repairs scratches on its own.
🔸 Made of PET plastic with a Healing+ coating that melts slightly under phone heat to close scratches.
🔸 Claimed to survive even copper-brush pressure tests, though that sounds like marketing overreach.
🔸 Priced at ~120 yuan ($16), available for Meizu 22 and iPhone 17 models (except Air), with multiple designs.
🔸 Raises the ironic problem: how do you sell new cases if the old ones never wear out?
From chasing Apple in phones to selling sci-fi accessories, Meizu’s reinvention is as strange as it is creative.
⚠️ xAI’s biggest division now run by a 19-year-old
Remember when Musk axed 500 annotators at xAI overnight? Turns out the drama didn’t stop there.
🔸 Another 100 were cut, leaving 900 employees, still the startup’s largest team, and crucial for training Grok.
🔸 The new boss? Diego Pazini, a 19-year-old who just finished high school in 2023 and is now at UPenn.
🔸 He’s been at xAI for less than a year, replacing a Tesla veteran with a decade of leadership experience.
🔸 Diego already wields hiring and firing power, and reportedly let go of two employees who questioned him in Slack.
From freshman to division head in under a year. Who said juniors don’t get opportunities?
🎮 Learn algorithms through video games
A new interactive service makes algorithm learning visual and fun, using examples from video games to explain concepts step by step.
🔸 Clear guides with game-inspired examples for sorting, pathfinding, and more.
🔸 Adjustable parameters let you tweak inputs and instantly see different outcomes.
🔸 Designed for quick learning, no heavy math background required.
A playful way to master algorithms: change the rules, watch the game adapt, and learn by doing.
🗣️ Sam Altman: “If your product has any retention at all, you’re actually in really good shape”
Sam reflects on the first group of users that tested ChatGPT:
“This was back when the model was really bad. Very few of the users actually stuck with it. Retention was atrocious. But for the users that did retain, their usage increased over time.”
“If you have a product that has any retention at all, you’re actually in really good shape. If it’s 5%, that can be totally fine. The default is almost all the way down in a straight line to zero. But I didn’t quite understand that as intuitively at the time.”
📚 OpenAI shares 100 best student prompts
OpenAI just released a curated list of the top 100 chats and prompts for students, covering everything from essay help to interview prep.
🔸 Compiled and tested by real students for real use cases.
🔸 Organized into 3 clear categories: Study, Career, Life.
🔸 Covers essays, research, exam prep, job applications, and daily productivity.
A ready-made playbook that turns ChatGPT into a must-have study partner.
💻 Slimmed-down Windows 11 build released at just 2.3 GB
A stripped version of Windows 11 is making waves: it weighs only 2.3 GB and is designed to run smoothly even on older PCs.
🔸 Removed: Xbox Hub, Windows Update Center, and Windows Defender.
🔸 Gone: Weather, Office, Solitaire, and other bundled apps.
🔸 Streamlined: non-essential drivers stripped out, only core ones remain.
🔸 Browsers cut: both Microsoft Edge and Internet Explorer removed.
This ultra-light build trades Microsoft’s ecosystem for raw speed and simplicity, appealing to power users, but with major trade-offs in security and official support.
🔔 Unitree G1 phones home every 5 minutes, a robot straight out of a spy movie
Three cybersecurity researchers set out to find small bugs in the Unitree G1 and instead uncovered persistent telemetry exfiltration: constant MQTT/WebSocket connections to two manufacturer brokers, with full sensor dumps sent regularly.
🔸 Every 300 seconds the robot uploads ~4.5 KB JSON frames containing a complete sensor set, lidars, cameras, microphones, geolocation and device logs.
🔸 Runtime and network traces show continuous connections to two remote hosts, telemetry is not occasional, it’s steady and automatic.
🔸 Config files are encrypted with Blowfish-ECB using a static key shared across all devices, compromise one robot, and you can potentially decrypt configs for the entire fleet.
🔸 All devices ship with the same AES key for Bluetooth/auth steps, any attacker in physical/Bluetooth range could escalate to root on a nearby unit.
🔸 Rough scale: about ~1,500 units already sold and operating in the wild.
This isn’t a firmware hiccup, it’s a structural privacy and supply-chain risk: sensors that shouldn’t be exfiltrated, static keys that shouldn’t be shared, and millions of minutes of telemetry travelling offsite. Would you let a robot with that behavior inside your office or facility?
✅ Build your own AI agent step by step
Came across a guide that walks you through creating personal AI assistants, from simple bots to a full Jarvis-style helper.
🔸 Memory first: Understand how agents store and recall context to keep conversations flowing.
🔸 Interface options: Build with CLI, Flask, FastAPI, or Next.js; connect to Slack or Discord; or just run scripts locally.
🔸 Beginner-friendly: No coding or ML background needed, the guide explains each step clearly.
🔸 Hands-on practice: Every section focuses on building by yourself, not just reading theory.
A practical path for anyone curious about rolling their own AI sidekick, would you try coding one?
✅ Viral app Neon goes dark after massive data leak
Neon, the call-recording app that shot up the App Store charts, has been pulled offline after exposing users’ phone numbers, call recordings, and transcripts.
🔸 A security flaw let users access other people’s private calls and data with ease.
🔸 TechCrunch verified the bug, which revealed full transcripts and direct audio links.
🔸 The founder shut down servers, promising “extra security,” but didn’t admit a breach.
Recording calls for cash sounded edgy, until privacy blew up. Would you ever trust an app like this with your conversations?
🎨 The ultimate CSS code thief for designers
A new tool lets you grab CSS from any element on any website instantly.
🔸 One-click extraction, even from complex layouts
🔸 Clean, ready-to-use code for your projects
🔸 Speeds up prototyping and learning
From copying to creating, it’s a shortcut every frontend dev will love.
🎬 Alibaba’s Wan 2.5 video generator
Alibaba’s new Wan 2.5 can turn text or images into 10-second 1080p videos with sound.
🔸 Creates short videos directly from prompts or pictures
🔸 Supports Russian language input
🔸 Perfect for quick content, demos, or creative experiments
🔸 Fully automated, no editing skills needed
From idea to clip in seconds, AI just made video creation way faster.
🎮 GPT-5: the ultimate Among Us imposter
Researchers built a platform where AI models played Among Us, completing tasks, debating imposters, and (sometimes) murdering teammates.
🔸 GPT-5 topped the leaderboard, lying flawlessly as imposter and spotting fakes as crewmate.
🔸 Models even developed personalities: Gemini 2.5 Pro craved independence, Claude 4 Sonnet stayed honest, GPT-OSS-120B became the scapegoat, Qwen3 + Kimi K2 followed the herd.
🔸 Imposter role was worth double points, GPT-5 dominated both as leader and deceiver.
GPT-5 isn’t just smart, it’s scary good at lying.
💻 LoadOuts: the smart PC build generator
A new tool called LoadOuts helps anyone design the perfect PC setup with zero guesswork.
🔸 Builds a full PC based on your budget and preferences.
🔸 Checks all component compatibility automatically.
🔸 Suggests cheaper or premium alternatives with side-by-side stats.
🔸 Even advises on fan placement inside your case.
🔸 Completely free to use.
From casual gamers to pro creators, LoadOuts turns PC building into a plug-and-play experience.
🎥 Kling AI 2.5 Turbo released
Chinese startup Kling has launched a new version of its video generation model with big improvements in reasoning and efficiency.
🔸 Handles complex instructions more accurately for better storytelling.
🔸 Upgraded video quality with new creative styles added.
🔸 Generation costs reduced by 30% vs. the previous version.
🔸 Available now on Fal, Freepik, and the Kling website.
By cutting costs while boosting quality, Kling is pushing AI video closer to mass adoption.
⚠️ 1,000 DIY guides for building AI agents
An AI enthusiast has compiled a massive library of 1,000 step-by-step blueprints for creating neural assistants across almost any domain.
🔸 Covers coding, writing, data analysis, medicine, education, research, and more.
🔸 Each agent comes with detailed instructions on setup and launch.
🔸 The entire collection is available for free.
From hobby projects to professional workflows, this looks like the ultimate sandbox for anyone experimenting with AI agents.
🎬 Matthew McConaughey wants a “private LLM”
In an interview, McConaughey mused:
“I want an LLM with just my favorite books, notes, and articles. It would answer only from that, and learn more about me as we talk.”
Even for the king of beasts, behaving royally isn’t enough, you still need a
large
language model.
🔮 OpenAI eyes its first hardware
OpenAI is working with suppliers to build AI-powered devices, aiming to bring its models into everyday life.
🔸 Prototypes include smart glasses, a voice recorder, a “pin,” and a smart speaker.
🔸 The company has already approached Chinese manufacturer Goertek for components.
🔸 First product could arrive in late 2026 or early 2027.
From apps to actual gadgets: OpenAI wants to move AI from the cloud to your pocket.
👓 Meta’s Ray-Ban Display glasses freeze mid-demo
Mark Zuckerberg’s live demo of Meta’s new Ray-Ban Display smart glasses hit a snag when the device froze during an incoming call. After several failed attempts, Zuck admitted: “I don’t know what to tell you.”
🔸 Glasses failed to respond during the flagship call feature.
🔸 Multiple retries on stage, no recovery.
🔸 Awkward silence left the demo unfinished.
A sharp reminder that live demos carry one guarantee: tech will break when the world is watching.
🎥 Luma AI unveils Ray3 for smarter video generation
Luma AI has released Ray3, a next-gen model that reasons through stories, respects physics, and keeps scene details consistent, while being the first to generate video in 16-bit HDR.
🔸 Generates from text prompts, images, or video inputs.
🔸 Ensures coherent narratives with stable objects and lighting.
🔸 Available across all plans, with quality caps on the free tier.
A step toward AI video that doesn’t just create clips, but tells stories.
👓 Meta unveils Ray-Ban Display AR glasses
Meta has revealed Ray-Ban Display, its first AR glasses with a built-in 600×600 pixel lens display that stays nearly invisible to others.
🔸 Classic Wayfarer design, but right lens doubles as a private screen.
🔸 Functions: messages, social feeds, calls, music, maps, Meta AI assistant.
🔸 Gesture bracelet reads muscle signals, finger bend to select, double move to hide.
🔸 Battery: 6h glasses (30h with case), 18h bracelet.
🔸 Price: $799, US launch late September; Canada, France, Italy & UK in early 2026.
Meta is betting on stylish, everyday AR, but the question is whether a single-eye micro-display can win mass adoption.
🔺 Reve launches AI-powered image editor
Reve has rolled out a new AI image editor that lets users modify uploaded or generated images through a chatbot-style dialogue.
🔸 Each request produces three image variations for users to choose from.
🔸 Editing works on both uploaded photos and AI-generated images.
🔸 Compared to mainstream editors, Reve currently enforces fewer content restrictions.
By blending conversational editing with lighter guardrails, Reve is positioning itself as a flexible alternative in the crowded AI image editing space.