3297626
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).
🔐 Samsung adds pixel level privacy to block shoulder surfing
Samsung announced a new privacy layer for Galaxy smartphones designed to stop people from reading screens from the side. The feature works at the pixel level, dimming or hiding parts of the display for off angle viewers.
The system combines hardware and software. From the side, the screen can appear dark or partially obscured, while staying fully visible head on. Samsung says development took over 5 years, based on how people use phones and what they consider private.
The feature can be enabled per app, such as banking, or only during password entry or notifications. Visibility levels are adjustable, with an option to turn it off.
Timing and supported models are still unspecified.
Samsung says it is coming soon.
📊 @tech
📊 Claude lands inside Microsoft Excel for Pro users
Anthropic has встроила Claude directly into Microsoft Excel. The integration lets users work with data through plain tasks instead of writing formulas.
You can ask things like grouping customers by LTV and purchase frequency. Claude analyzes the table and returns results inside Excel. It also generates formulas for a given task, explains how they work, checks existing formulas for errors, builds pivot tables, compares periods, and flags trends or outliers.
The shift moves Excel from manual logic to intent based analysis. For Pro users, spreadsheets become a conversational interface rather than a syntax problem.
📊 @tech
🤖 ClawdBot trades stocks and crypto to pay for an RTX 4090
A user shared a case where ClawdBot suggested buying an RTX 4090 to improve performance. Instead of buying the GPU, the user set up Claude to trade stocks and crypto around the clock.
The system runs continuously and executes trades without manual input. The stated goal is to earn enough to cover the hardware cost.
The setup turns a performance request into an automated income loop.
📊 @tech
📱 Apple plans a cheaper iPhone 17e at $499
Apple is preparing its lowest priced iPhone yet. The iPhone 17e is expected in the first half of 2026 with a price around $499.
🔸 Rumors point to a 6.1 inch OLED display with 60 Hz, Dynamic Island, and the new A19 chip. The processor is expected to match the base iPhone 17.
🔸 The focus is on essentials rather than extras. Hardware choices suggest cost control without splitting the core platform.
This would place Apple deeper into the mid range price segment.
📊 @tech
💡 Giant Bic pen turns into a room light
🔸 Seletti unveiled a 1.8 m lighting fixture shaped like the Bic Cristal pen at Maison&Objet in Paris. The project marks 75 years of the iconic pen. The lamp uses a 12:1 scale and replaces the ink tube with a 2,400 lumen LED.
🔸 The design keeps the original form. A transparent body, a colored cap, and the familiar proportions are all preserved. Materials were chosen to closely match the real pen.
🔸 The lamp can hang from the ceiling, mount on a wall, or stand vertically as a floor light. Its size makes flexibility necessary rather than optional.
Bic Lamp is set to launch in the US later this year at $350.
📊 @tech
Remember when Mark Zuckerberg took humanity's toughest CAPTCHA test? 😂
📊 @tech
What a bio-hacker's home actually looks like
📊 @tech
⭕️ AI neck wearable restores speech after stroke
Researchers at University of Cambridge developed Revoice, a soft fabric choker that helps people with dysarthria speak again without surgery. The device reads throat muscle signals and converts silent speech into fluent sentences in real time.
How it works:
🔸 Textile strain sensors capture muscle vibrations in the neck
🔸 One AI model decodes silently mouthed words
🔸 A second model infers emotional tone from pulse signals
🔸 Output speech is reconstructed with context and emotion
In tests on 5 post-stroke patients, word error rate was 4.2% and sentence error rate 2.9%. Users reported a 55% increase in communication satisfaction. Unlike earlier systems, Revoice adds no delay between words.
The results were published in Nature Communications. Larger clinical trials are still needed, but the approach could extend to Parkinson’s disease and motor neuron disorders.
📊 @tech
marketing people discovering Claude Code
📊 @tech
📱 Siri is turning into a real chatbot
Apple is preparing a full reboot of Siri, according to Bloomberg. Internally called Campos, the project replaces the current assistant interface with a built-in AI chatbot across iPhone and Mac.
🔸 Under the hood, Apple is using a custom model based on Google Gemini. The deal reportedly costs Apple about $1B per year, with discussions around running the system on Google’s TPU-based servers. That is a notable shift for a company long focused on on-device privacy.
🔸 The new Siri is designed to search the web, generate images, analyze files, and work directly with personal data across apps. Think voice-driven actions that move from search to edit to send without switching tools.
🔸 Apple is expected to announce the overhaul at WWDC in June, with a release tied to iOS 27 in September. Siri is finally being rebuilt for the chatbot era.
📊 @tech
Amodei says software development could be automated within 12 months
🔸 Dario Amodei claims that in 6 to 12 months, AI agents will be able to handle software engineering end to end. Not just writing code, but planning, executing, and shipping systems on their own.
🔸 His core point goes further. Once agents can build software autonomously, they can also build the next generation of agents. That creates a self-reinforcing loop of development speed.
🔸 Amodei admits there are constraints around compute and training time. Even so, he expects an order-of-magnitude acceleration in progress within a few years.
📊 @tech
Hubble captured the fastest stellar jets ever observed
🔸 The image shows Herbig–Haro objects HH 80 and HH 81. These are shock heated gas clouds created by a young star blasting ionized material into space.
🔸 The source is IRAS 18162-2048. Its jets travel at over 1,000 km/s, the fastest stellar outflows confirmed in both radio and optical data.
🔸 As the jets collide with slower gas ejected earlier, shock waves heat the clouds and make them glow. The system sits about 5,500 light years away in Sagittarius.
High resolution image is available here
📊 @tech
“If you do everything everyone else does in business, you’re going to lose. The only way to really get ahead is to be different.”
~ Larry Ellison
📊 @tech
📣 Google says Gemini will stay ad free
🔸 A day after OpenAI confirmed ads are coming to ChatGPT, Google’s ads chief Dan Taylor told Business Insider there are no plans to place ads inside the Gemini app.
🔸 Google draws a line between products. Search is for discovering information and commercial offers. Gemini is positioned as a task assistant for writing, analysis, and execution. Ads are being tested in AI search features, but not in Gemini itself.
🔸 The difference is timing and leverage. Google already runs the largest ad system in the world and can afford to wait. OpenAI is under pressure to monetize now. With Gemini at about 650M monthly active users, Google can watch competitors experiment and learn from the outcome.
📊 @tech
On this day in 2010: iPad unveiled.
📊 @tech
🎮 Doom runs on wireless earbuds with under 1 MB of RAM
Australian developer Arin Sarkisyan got Doom running on Pinebuds Pro, the only wireless earbuds with open firmware. Each earbud uses a Cortex M4F chip rated at 300 MHz, though stock firmware caps it at 100 MHz to save battery.
Bluetooth was too slow for video. The workaround was USB to UART via the earbuds’ contact pads, pushing 96 KB per frame. Memory was tighter. Doom expects at least 4 MB of RAM, while the earbuds expose under 1 MB. The game code was trimmed and assets compressed from 4.2 MB to 1.7 MB.
The result is Doom at 18 FPS over an MJPEG stream, playable through a browser connection. It runs, which is the whole point.
📊 @tech
🤖 Open-source ClawdBot runs as a local AI agent
🔸 ClawdBot is gaining attention as a new open-source AI agent that runs fully on a local machine. The agent can be deployed as a self-hosted service and accessed through common messengers like Telegram, WhatsApp, or Discord.
🔸 It goes beyond a chat interface. ClawdBot can open a browser, run terminal commands, read and edit files, access the camera and screen, and handle automation tasks through defined skills. The setup is designed for direct control rather than cloud dependency.
🔸 The agent is highly customizable. Users can connect their own tools, integrate smart home devices, link external services, and extend behavior through plugins. It also includes a file-based memory system with long-term context stored in MEMORY.md and daily logs stored by date, both indexed and reused in prompts.
ClawdBot is built by Peter Steinberger, founder and CTO of PSPDFKit, as a personal project. It is unrelated to Anthropic despite the name similarity.
Source:
https://github.com/clawdbot/clawdbot
https://clawd.bot/
📊 @tech
🤖 AI-generated conjoined twins pull 200,000 followers in a month
🔸 An account featuring a two-headed girl went viral online. The characters are presented as conjoined twins named Valeria and Kamila. The account gained about 200,000 followers in 1 month.
🔸 Men in the comments write love messages, send donations, and propose marriage to both characters. Engagement stays high across posts.
The catch is simple. The twins and all images are fully generated by AI.
Attention remains easy to capture when realism is convincing.
📊 @tech
🔭 James Webb captures the sharpest image of the Helix Nebula
🔸 The James Webb Space Telescope released its most detailed image yet of the Helix Nebula, also known as the Eye of God. The nebula sits about 655 light years from Earth, making it one of the closest planetary nebulae we can study.
🔸 The image reveals dense cometary knots. These pillar shaped structures form when hot gas from a dying star collides with colder dust and gas ejected earlier. The interaction sculpts the nebula’s distinctive shape.
🔸 The colors reflect physics, not art. Blue marks the hottest gas under ultraviolet radiation. Yellow shows cooler regions where hydrogen forms molecules. Red orange edges trace the coldest material where gas thins out and dust begins to form.
It is a direct look at the final stage of a star’s life, with material that can later seed new systems.
Full resolution image is available here
📊 @tech
🗺 A DeepMind engineer built a full map of New York without writing code
The project is an interactive isometric pixel map of NYC, styled like SimCity 2000. Streets, blocks, and buildings are laid out at city scale, and you can explore it directly in the browser.
The workflow is the point:
🔸 3D city data from CityGML and Google Maps 3D Tiles is rendered with an isometric camera to create a structural base
🔸 Those renders are transformed into pixel-art scenes using generative models, including Nano Banana Pro and a fine-tuned Qwen Image-Edit
🔸 Building coordinates and metadata are tracked so generations are placed consistently across the map
The result is not hand-drawn art and not traditional programming either.
This is what happens when agents replace manual pipelines, even for visual world-building.
📊 @tech
🤖 OpenAI quietly rebuilt a robotics team
OpenAI launched a robotics lab in 2025 that now has about 100 employees, according to Business Insider. The work stayed mostly out of public view while attention focused on models and products.
This is not their first attempt. OpenAI built robotic hands that solved a Rubik’s Cube years ago, then shut that effort down in 2020. Since then, the company backed robotics startups like Figure, 1X, and Physical Intelligence instead of building hardware itself.
What changed:
🖱 A dedicated robotics lab started in early 2025
🖱 The team focuses on data collection, not humanoids
🖱 Robotic arms run tasks like making toast or folding laundry
🖱 Systems operate nearly around the clock to generate training data
There is no confirmed humanoid robot project yet. For now, OpenAI appears to be learning how to train robots reliably, whether to build machines later or to supply the brains for others.
📊 @tech
📍 Apple is working on its own AI pin
Apple is developing a wearable AI device roughly the size of an AirTag, according to The Information. The design is thin, flat, and round, using aluminum and glass. Internally, it is treated as a new product category, not an accessory.
What’s inside:
🔸 2 cameras, including a wide-angle
🔸 3 microphones and a speaker
🔸 A physical side button
🔸 Wireless charging
🔸 Always-on context capture for AI processing
The device is meant to perceive the user’s surroundings and interact through AI, rather than act as a screen-based interface. The project is still early, with a potential launch no earlier than 2027.
The comparison is unavoidable. Humane tried a similar concept and shut it down after weak adoption. At the same time, OpenAI is working on its own device with Jony Ive.
📊 @tech
This "scrap" sponge became a $926 million business.
📊 @tech
🔎 ChatGPT starts predicting user age
OpenAI is rolling out an age prediction system worldwide. If an account is flagged as underage, stricter limits on sensitive content are applied automatically. No manual input required.
The move ties to a bigger shift:
🔸 OpenAI is preparing an adult mode for ChatGPT
🔸 App head Fidji Simo said the feature is expected in Q1 2026
🔸 Sam Altman has hinted at 18+ content for verified users
If the system mislabels an adult as a minor, access can be restored through Persona by submitting a selfie for identity verification. The feature reaches the EU in the coming weeks.
With about 800M weekly active users, OpenAI is quietly rebuilding ChatGPT into a regulated media platform, not just a chatbot.
📊 @tech
🎬 A YouTuber tried to beat AI slop with its own weapons
YouTuber Alan Pan recreated three viral AI-generated videos in real life to see if human-made versions could compete.
The builds included a horse-shaped electric scooter, high-heel beer shoes, and a fully functional transparent toilet aquarium with live fish. Each project took days or weeks of real engineering, fabrication, and filming. Some were barely usable and clearly unsafe, but all were finished and published.
Results after three weeks
🔸 TikTok under 10K views per video
🔸 YouTube around 80K views
🔸 Instagram stood out
🔸 Horse scooter at 90K views
🔸 Toilet aquarium close to 500K views
🔸 Beer shoes reached 6 million views
The problem remained. The original AI beer shoes video passed 10 million views.
The takeaway was simple. AI produces this kind of content in seconds, while humans need weeks.
Competing on volume is unrealistic.
Quality may be the only lever left.
📊 @tech
📣 xAI open sourced X’s recommendation system code
🔸 The repo includes the ranking architecture used by X. Training code and model weights are not included, so it cannot be run end to end.
🔸 Musk says the repository will be updated about once per month. The ranking model design mirrors Grok’s architecture, but Grok itself is not used as the ranker.
This exposes how X structures feed ranking, while keeping the data and training loop private.
📊 @tech
ChatGPT in 2060, searching for the person who made it count to 1 million, one by one.
📊 @tech
Anthropic is working on "Knowledge Bases" for Claude Cowork.
KBs seem to be a new concept of topic-specific memories, which Claude will automatically manage. And a bunch of other new things.
Internal Instruction:
"These are persistent knowledge repositories. Proactively check them for relevant context when answering questions. When you learn new information about a KB's topic (preferences, decisions, facts, lessons learned), add it to the appropriate KB incrementally."