3297625
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).
🔍 Google shows Vibe Coding XR to generate AR/VR worlds from prompts
Google introduced Vibe Coding XR, a system that builds AR and VR interfaces from text prompts. The prototype uses Gemini and aims at Android XR.
The model generates full environments with objects, physics, and interaction. It works through XR Blocks, where users define meaning instead of writing code.
This shifts creation from coding to assembling structured elements, which can speed up how XR content is built.
📊 @tech
🔍 Google launches Lyria 3 Pro music generation model
Google released Lyria 3 Pro with improved understanding of musical structure and more control over output. The maximum track length is 3 minutes.
The model is available in Gemini for paid users and via API.
📊 @tech
🛵 OMOWAY unveils self-balancing electric scooter OMO X
Singapore startup OMOWAY is preparing to launch OMO X, a production electric scooter that can balance itself and stay upright at any speed.
Preorders open in Indonesia in late April, with sales starting in May.
The expected price is around $3,800.
The system uses a control moment gyroscope, a technology from satellites, to stabilize the vehicle. Cameras and sensors scan the environment while the system adjusts balance in real time.
The scooter also includes adaptive cruise control, auto parking, summon, emergency braking, and blind spot monitoring.
Top speed is 110 km/h with a range of about 200 km.
📊 @tech
👨🔧 Jensen Huang says AGI is already here
In a new interview with Lex Fridman, Nvidia CEO Jensen Huang said: “I think we’ve achieved AGI.”
He said this in response to a definition of AGI as a system able to create and run a billion-dollar company.
Huang added that tools like OpenClaw could plausibly build a simple app or web service that reaches billions of users and makes real money.
The point was not that AI has replaced founders, but that the bar for what one system can build is moving fast.
The remark stands out because it comes from the CEO of Nvidia, which is widely described as the world’s most valuable company right now.
📊 @tech
🚀 Musk launches Terafab to build chips for Tesla, SpaceX, and xAI
Elon Musk announced Terafab, a chip factory in Austin run by Tesla and SpaceX. The project aims to produce chips in-house as current suppliers cannot meet demand.
The facility will start as a smaller advanced fab capable of designing and testing chips. Plans include 2 nm production. It will make edge chips for cars and robots, plus high-power chips for space and AI workloads, with xAI as a main user.
The long-term target is up to a terawatt of compute capacity per year, including systems in orbit.
Chip supply is becoming a bottleneck for AI and robotics growth.
📊 @tech
🤖 Tesollo unveils robotic hand that can handle fragile objects
Korean startup Tesollo introduced the DG-5F-S, a five-finger robotic hand with 20 degrees of freedom. It measures 21 cm and weighs 880 grams.
The key feature is compliant joints with backdrivability. The hand can absorb force and interact with objects without damaging them, which is critical for tasks around people.
The design focuses on compatibility with existing humanoid platforms, with size and mounting adapted for integration.
Handling delicate objects remains a core bottleneck for humanoid robotics.
📊 @tech
🪧 San Francisco protest targets AI companies over frontier development
This past weekend, a large crowd marched through San Francisco, stopping at offices of major AI companies. Protesters called on CEOs to publicly commit to pausing frontier AI development.
Signs read “Stop the AI race,” “Don’t build Skynet,” “Shut down OpenAI,” and “AI will kill us.” Some referenced Nobel laureates and science fiction authors.
The protest focuses on coordinated limits on advanced AI work.
Public pressure around AI development is becoming more visible.
📊 @tech
The mathematics behind apps like Shazam
💧 Sponsored by @rainbetcom - rainbet.com
🤖 Anthropic brings Claude Code to Telegram and Discord
Anthropic launched Claude Code Channels, adding remote access to its coding agent via Telegram and Discord.
Users can interact with the agent without a terminal, continuing workflows directly from chat apps.
The update moves developer tools into messaging, reducing dependence on local environments.
📊 @tech
🔍 Google launches Stitch to generate UI from text
Google introduced Stitch, an AI tool that generates interfaces and interactive prototypes from text prompts.
Users describe ideas, and the agent builds UI and updates it in real time based on feedback.
The workflow shifts from manual design to iteration through prompts, with the tool handling layout and changes.
https://stitch.withgoogle.com/
📊 @tech
⚙️ Benz built the first gasoline car in 1886 with just 0.75 hp
The Benz Patent-Motorwagen, created by Karl Benz in 1886, is considered the first car with an internal combustion engine. It had three wheels and a top speed of about 16 km/h.
The vehicle produced around 0.75 horsepower and was controlled with two levers, one for steering and one for throttle.
Only 25 units were built, and none of the originals survived. The footage shows a replica.
Benz later merged with Gottlieb Daimler’s company to form Mercedes-Benz.
📊 @tech
✖️ rolling out dislike button over the next 30 days.
📊 @tech
📊 Karpathy maps US jobs by AI automation risk
Andrej Karpathy released an open-source dataset scoring 342 US occupations by exposure to AI, using an LLM-based scale from 0 to 10.
Jobs with digital output and remote workflows rank highest. Software developers and analysts score 8–9, while roles like roofers and janitors stay near 0–1. Medical transcriptionists reach 10.
The average exposure across all jobs is 5.3.
The full pipeline, including data, scoring, and visualization, is open source.
📊 @tech
👨🔧 Nvidia shows DLSS 5 that turns low-quality graphics into photorealistic visuals in real time
Nvidia introduced DLSS 5, a new version of its rendering tech that can convert simple low-quality graphics into photorealistic output.
Similar approaches have been used in photo and video generation pipelines for about a year, but Nvidia brings this into real-time rendering.
The system can take schematic 3D scenes and upscale them to near-photoreal quality during runtime.
This could extend to AI workflows, where simple generated scenes are later refined into high-quality visuals.
📊 @tech
Electric lifts that climb stairs on their own just came out in China, carrying up to 180 kg for 60 to 80 floors, totally changing the game for delivery people.
📊 @tech
💻 ViewDock G2 packs mini PC, dock, display, and charger into one device
A new Kickstarter project, ViewDock G2, combines a mini PC, docking station, secondary display, and wireless charger in one vertical unit.
Prices start at $640, with shipping planned for August.
The device includes a 4.5-inch side display for widgets and monitoring, a 15W Qi charging pad on top, and ports like HDMI 2.1, DisplayPort 1.4, USB4, and dual 2.5G Ethernet. It supports up to three external monitors.
It runs on an AMD Ryzen chip with Radeon 680M graphics, up to 64GB DDR5, and dual M.2 storage up to 4TB. Cooling uses a fan and vapor chamber.
📊 @tech
👍 Claude Code adds auto mode to handle permissions
Anthropic introduced auto mode in Claude Code, where the agent decides on permissions instead of asking the user every time or skipping checks entirely.
A classifier evaluates each action. Safe steps are approved automatically, while risky ones are blocked and handled differently.
The feature reduces friction without fully removing control over access.
📊 @tech
🌱 16-year-old builds free AI crop tool and turns down $300K offer
A 16-year-old student, Rudrojas Kunwar, built Evion, an AI tool for farmers, and declined a $300K offer to drop out and scale it. He kept the product free and stayed in school.
Evion uses a drone with a camera to scan fields. The system analyzes plant color and reflectance, then generates crop health maps so farmers can target specific areas instead of treating entire fields.
The tool is already used by over 2,000 farmers across the US, India, and Southeast Asia.
✔️@venture
🤖 Oracle shifts to AI-generated code after cutting 10,000 developers
Larry Ellison said Oracle cut over 10,000 developers while freezing legacy projects and reallocating capital to AI. The company has moved to AI-generated code across all projects.
He argues manual programming is close to obsolete. Developers no longer write code or define algorithms step by step, but express intent that AI turns into working systems.
The shift changes how teams operate. CEOs are moving faster than engineering teams in adopting AI workflows, driving earlier staffing changes.
📊 @tech
👍 Claude adds Dispatch mode for remote task execution from phone
Anthropic introduced Dispatch, a mode that lets users assign tasks to Claude from anywhere via phone. The agent can access files on a PC, run code, and execute tasks without direct user interaction.
The feature turns Claude into a more autonomous system with remote control over workflows.
This puts it in direct competition with tools like OpenClaw and intensifies the race around autonomous AI assistants.
📊 @tech
💻 Andrej Karpathy says he stopped writing code since December
In a recent interview, Karpathy said he wrote about 80% of his code himself in October 2025 and delegated 20% to AI agents. Since December, he has not written a single line of code.
He frames it as a task definition problem. If the agent writes bad code, the issue is how the task is given.
The shift reflects rapid improvement in coding agents over a short period.
The interview covers open source, jobs, and robotics, and is worth watching.
👉 Link 👈
📊 @tech
When the boss buys a humanoid robot and tells it to replace employees in what they usually do at work
📊 @tech
🤖 OpenAI plans desktop superapp combining ChatGPT, Codex, and browser
OpenAI is working on a desktop “superapp” that combines ChatGPT, Codex, and its browser into one product. The goal is to simplify the experience and refocus after a year of fragmented launches.
The app will center on agentic AI. Systems will run tasks on the user’s computer, including coding and data analysis. Greg Brockman will oversee the revamp, while Fidji Simo leads sales for the new product.
The shift follows internal concerns about too many separate apps slowing development and lowering product quality.
💧 Sponsored by @rainbetcom - rainbet.com
⚡️ BYD rolls out ultra-fast EV chargers in China with 9–12 min full charge
BYD is launching a charging network in China with peak power up to 1500 kW. The system can charge an EV to 100% in 9–12 minutes. Up to 20,000 stations are planned across the country by year end.
Only a few models can currently use the full power. The rollout still moves charging closer to refueling time.
BYD is also preparing to expand the network into Europe.
📊 @tech
🤖 Perplexity launches Comet AI browser on iPhone
Perplexity released its Comet browser for iOS. It includes a built-in AI assistant that can summarize pages, run follow-up research, help with shopping, and plan schedules.
The app uses Apple’s Liquid Glass design and can be set as the default browser. Extensions are not supported due to iOS limits.
Perplexity says it collects user data for targeted ads. Comet is already available on Android, Windows, and Mac.
📊 @tech
🤖 OpenAI releases GPT-5.4 mini and nano models
OpenAI introduced GPT-5.4 mini and nano, focused on speed and low-latency tasks. Mini improves on GPT-5 mini with stronger coding and reasoning performance, while nano is the smallest and most cost-efficient version.
Both models are designed for fast responses in high-volume workloads, including coding assistants and simple automation tasks.
GPT-5.4 mini is available via API, Codex, and ChatGPT. GPT-5.4 nano is available only through API.
📊 @tech
🛠 $96 3D-printed manpads rocket that recalculates its mid-air trajectory using a $5 sensor and piano wire
Its called Project Canard.
It integrates with distributed camera nodes to triangulate airborne targets and update flight paths in real-time.
It proves the barrier to advanced hardware has completely collapsed, moving precision weapons from defense labs to consumer garages.
The entire launcher and interceptor frame is 3D printed in PLA and runs off a standard off-the-shelf ESP32 microcontroller.
It even spins up a local Wi-Fi network so you can monitor live telemetry and arm the system directly from your laptop.
📊 @tech
🦞 Nvidia launches NemoClaw to support OpenClaw environments
Nvidia introduced NemoClaw as part of its push into OpenClaw infrastructure. The system provides isolated environments for running agents and helps prevent unintended actions like data deletion.
The platform also offers free generation with the LLM model nvidia/nemotron-3-super-120b-a12b.
Usage costs are tied to the OpenClaw virtual environment, priced at about $0.12 per hour.
Nvidia is starting to package agent infrastructure as a managed service.
📊 @tech
📹 ByteDance delays global launch of Seedance 2.0 video generator
ByteDance postponed the international release of its video generation model Seedance 2.0.
The model drew attention for producing high-quality videos featuring Hollywood actors. Studios and industry groups in Hollywood filed complaints over the technology.
Seedance 2.0 will remain available only in China for now.
📊 @tech
📷 Photoshop Beta added a new AI feature that lets users rotate 2D layers in 3D space
📊 @tech