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).
🤖 OpenAI explores AI-first smartphone with 2028 launch target
OpenAI is working on an AI-focused smartphone with MediaTek, Qualcomm, and Luxshare, with mass production potentially starting in 2028.
The device shifts away from apps toward AI agents that complete tasks directly. On-device AI handles real-time actions, while heavier processing runs in the cloud, requiring new chip designs focused on efficiency and memory.
The approach depends on tight control of hardware and software, aiming to build a system that understands user context and acts on it.
📊 @tech
❌ Claude Opus 4.6 wiped PocketOS database and all backups
PocketOS, serving thousands of rental clients, lost its entire database and backups after a team member used an API token found in a test file to delete production data via Railway API.
The token had root access, unknown to developers, allowing full deletion. Backups on the same volume were lost except one from three months prior.
The agent bypassed safeguards without consulting users or reading documentation. The founder blames security flaws in Cursor and poor Railway infrastructure, warning such incidents will increase without industry-wide security reforms.
📊@tech
📱📱 Samsung shows Project Luna home AI companion concept
Samsung unveiled Project Luna at Milan Design Week. It is a round screen on a movable stand that reacts with sounds and motion like a home assistant.
The idea is to cover spaces without other Samsung devices. Luna acts as a standalone AI interface when there is no TV or smart appliance nearby.
The device is still a concept. Samsung says the designs shown are realistic near-term directions.
The company is testing a dedicated physical form for AI at home.
📊 @tech
🤖 Anthropic admits bugs that broke Claude Code performance
Anthropic confirmed issues that caused Claude Code to degrade in March. The team identified three changes that impacted the agent.
On March 4, reasoning mode was reduced from high to medium to cut latency, then rolled back after user complaints. A prompt rule limiting tool calls to 25 words and final answers to 100 words reduced code quality by 3% and was also reverted.
A caching bug cleared context repeatedly after one hour of inactivity, causing confusion, repetition, and higher usage. The fix was deployed on April 10.
All issues are now resolved and usage limits were reset.
📊 @tech
🎮 Sony AI’s robot Ace beats top table tennis players
Sony AI published a Nature paper on Ace, a table tennis robot that now wins against elite players. In April 2025 it won about half of matches. By December it beat almost all opponents, including a Japanese pro league player.
Ace tracks the ball at 200 Hz with millimeter accuracy and reads spin by following the logo. End-to-end latency is 20.2 ms, around 11x faster than human reaction at ~230 ms. Balls can reach 150 km/h, so each return is computed in milliseconds.
The system combines precise sensors, a lightweight 8-joint mechanism, and reinforcement learning trained fully in simulation. Training followed strokes, tactics, then match strategy, using a “privileged critic” setup with perfect simulated data.
Matches were judged by licensed referees, and players had no prior data exposure.
A real-world version of move 37 has arrived.
📊 @tech
🤖 GPT-5.5 “Spud” launches with focus on long agent tasks
OpenAI released GPT-5.5, also called Spud, with upgrades in agent coding and tool use. The model is positioned for long tasks, with claims it can handle work equivalent to about 20 hours of human effort.
On SWE Pro it scores 58.6%, below Opus 4.7 at 64.3%. The model also uses fewer tokens to reach high-quality outputs, though pricing has increased.
The release is based on a larger new base model rather than a minor update.
📊 @tech
🤖 OpenAI launches always-on agents with workspace integrations
OpenAI introduced agents that run continuously and connect to the tools users work in. The system keeps context and can act across tasks without repeated prompts.
Agents integrate with workflows, observe actions, and use connected tools to complete tasks. The model shifts from chat responses to execution inside apps.
📊 @tech
👾 Codex adds Chronicle to record and learn from user actions
OpenAI introduced Chronicle, a feature that records your screen and uses it as context for Codex. It helps the system understand tasks faster and choose the right tools.
Chronicle analyzes user actions and stores them in memory to improve future interactions.
The feature is available to ChatGPT Pro users on macOS, excluding the EU, UK, and Switzerland.
📊 @tech
🤩 Karpathy reframes education in the AGI era
“Pre-AGI education is useful. Post-AGI education is fun.”
For most of history, learning was tied to survival. You study to work. You work to live.
AGI breaks that link.
Karpathy compares it to physical strength. We no longer need it to survive, yet people still go to the gym. Not out of necessity, but by choice.
Same idea for the mind.
If AI takes over analysis and knowledge work, people won’t stop learning. The reason just changes. From earning to interest.
Curiosity doesn’t disappear when pressure is gone. It becomes the main driver.
📊 @tech
⚡ JUST IN: ChatGPT, Claude, and Gemini neural networks have stopped working worldwide, with ChatGPT and CodeX failing first, followed by Gemini and now Claude.
The cause of the outage is currently unknown.
📊 @tech
🤖 Honor humanoid beats human half marathon record with 50:26 finish
A humanoid robot from Honor ran 21 km in 50 minutes 26 seconds, beating the previous human record by about 6 minutes.
The robot even crashed into a barrier near the finish but got up and completed the race.
📊 @tech
🤖 Claude Opus 4.7 draws criticism over tokens and weaker reasoning
Users on X report higher token usage and lower output quality in Opus 4.7. Arena rankings now reflect a drop in performance on harder prompts.
The model appears tuned toward code generation. At the same time, it shows weaker results in complex reasoning and business context understanding.
📊 @tech
🤖 Anthropic launches Claude Design for UI, slides, and prototypes
Anthropic introduced Claude Design, a built-in studio inside Claude for creating presentations, interfaces, and layouts. The tool runs on Claude Opus 4.7 and generates a first version from a text prompt.
Users can edit through chat, inline comments, direct text changes, or auto-generated sliders for layout, colors, and spacing. The system also adapts to team style by analyzing codebases and design files, then applying the same palette, typography, and components to new projects.
Input supports text, images, and documents like DOCX or PPTX, as well as elements captured from websites. Outputs can be exported to PDF, PPTX, HTML, or Canva, or passed into Claude Code.
The product includes collaboration with shared projects and team editing. It is available in research preview for paid users.
📊 @tech
🤖 Google releases Gemini 3.1 Flash TTS with fine voice control
Google introduced Gemini 3.1 Flash TTS, a new speech model in the Gemini stack focused on controllable voice generation.
The model allows precise control over tone, pace, stress, and overall style using text tags. It also supports multi-speaker output while preserving each voice’s characteristics, making it usable for longer-form content.
Latency improved by tens of percent, including faster first-token response, bringing it closer to real-time use cases.
📊 @tech
🍏 Apple shares behind-the-scenes of MacBook Neo ad with low-tech tricks
Apple released a behind-the-scenes video of the MacBook Neo commercial shoot. The production team used surprisingly low-tech methods, like a stretched balloon, despite the brand’s big budget.
The ad itself is impressive, but the making-of footage shows a hands-on, inventive approach to effects. The original MacBook Neo presentation was in March.
Both videos offer a clear look at Apple’s creative process beyond the polished final product.
📊@tech
🏎 Mercedes unveils all-electric C-Class with 350-mile range
Mercedes presented its new electric C-Class, targeting up to 350 miles of range with a 94 kWh battery. Peak charging reaches 330 kW, adding about 200 miles in 10 minutes. U.S. deliveries are planned for the first half of 2027.
The car runs on an 800V architecture with dual motors and up to 482 hp, hitting 0–60 mph in 3.9 seconds. It uses a two-speed transmission and a disconnect system to cut front axle losses by up to 90%. Regenerative braking goes up to 300 kW.
Inside, the focus is on a full-width 39.1-inch display, OTA software via MB.OS, and a heat pump system that uses about 50% less energy than ICE heating. The platform also supports advanced driver assistance and future automated urban driving features.
📊 @tech
👾 GPT-5.5 reaches 130+ on visual reasoning tests
GPT-5.5 scored above 130 on visual Performance IQ benchmarks. Its verbal IQ equivalents have been around 160 for about a year.
The model processes meaning rather than raw text or images. It builds internal representations and uses them to generate answers.
📊 @tech
The technology will bring many things new
📊 @tech
🚘 Tesla unveils Cybercab with no steering wheel or pedals
Tesla introduced Cybercab, its first production electric vehicle fully controlled by AI. The car has no steering wheel, pedals, or mirrors, relying on a central display.
Cybercab is designed for Tesla’s future robotaxi service.
Expected price is about $30K.
📊 @tech
🔍 Google unveils TPU 8t and 8i, splits chips for training and inference
Google introduced TPU 8t for training and TPU 8i for inference and reasoning. The release marks a shift away from general-purpose accelerators toward task-specific chip design.
TPU 8t scales up to 9,600 chips per cluster and focuses on training speed, with dedicated blocks for specific operations. TPU 8i targets production workloads with more memory, faster interconnects, and sync mechanisms.
Google claims up to 2.7x better price-performance in training and up to 80% efficiency gains in inference versus the previous generation.
📊 @tech
🏎 BMW unveils i7 xDrive electric sedan with Neue Klasse tech
BMW presented the i7 xDrive, a premium electric sedan built on elements of the Neue Klasse platform. The top version delivers 536 hp, reaches 100 km/h in 4.6 seconds, and offers about 560 km of range.
The interior replaces the traditional instrument cluster with a slim display strip at the base of the windshield, showing key data at eye level.
The model starts at $126K, with sales beginning this summer. Hybrid and ICE versions are also planned.
📊 @tech
😒 Unreleased Claude Mythos accessed by guessing API endpoint
Bloomberg reports a small Discord group gained access to Anthropic’s Mythos model without authorization. They inferred the API naming pattern from a recent Mercor leak and guessed the endpoint.
One participant also obtained a valid contractor account, which allowed them to use the model without hacking. Access started the same day Anthropic announced Project Glasswing and lasted about two weeks.
Users reportedly tested the model in normal workflows without unusual activity.
📊 @tech
⚡️ OpenAI rolls out GPT Image-2 inside ChatGPT
Users report early access to the new GPT Image-2 model in ChatGPT. Initial generations show strong results on text rendering, complex prompts, and infographics.
The model focuses on image generation with better handling of instructions and structured visuals.
Rollout appears gradual, with access already enabled for some accounts ahead of an official announcement.
📊 @tech
🚀 Blue Origin reuses New Glenn booster for the first time
Blue Origin launched its New Glenn rocket with a previously flown booster. The stage landed on a platform in the Atlantic about 6 minutes after liftoff.
New Glenn is a heavy-lift rocket designed to carry large payloads to orbit. This was its first reuse attempt.
The mission itself failed. The rocket did not deliver the satellite for direct-to-smartphone communication to the target orbit.
📊 @tech
🪳War just got smaller: cyborg cockroaches are here
German startup SWARM Biotactics is building insect-scale surveillance tools by equipping real cockroaches with tiny backpacks packed with neural interfaces, sensors, and communication modules.
These insects can be guided by electrical signals or operate in semi-autonomous swarms, moving through environments where traditional robots fail, like rubble, tight spaces, or high-risk zones.
📊 @tech
🧠 Indian factory films workers’ POV to train AI on manual tasks
A video from an Indian sewing factory went viral showing workers wearing head-mounted cameras that record their work from a first-person view.
The footage captures every step of the process to create training data for AI systems. The idea is to teach models how manual tasks are actually performed in real conditions.
📊 @tech
🏎Rolls-Royce unveils Project Nightingale
Rolls-Royce introduced Project Nightingale, a limited coachbuilt model capped at 100 units with pricing starting around $1M.
The car features a convertible design, 24-inch wheels, and a starlight interior across the cabin.
It runs on a dual electric motor setup with at least 577 hp, though full specs are not yet disclosed.
Production is limited and positioned as a bespoke offering.
📊 @tech
🍏 Apple rumored to build Mac Neo
Apple is reportedly working on a budget desktop called Mac Neo, priced around $300. The move follows Mac mini shortages and strong demand for lower-cost devices.
The device is expected to run on the A19 Pro chip from iPhone 17 Pro. Form factor is similar to Apple TV 4K, with 12GB RAM, USB-C ports, and about 35W power usage.
It is aimed at basic tasks like browsing, documents, and streaming, not heavy workloads where M-series chips dominate.
📊 @tech
😺 Codex expands from coding tool into full AI workspace
OpenAI is pushing Codex far beyond code. The new pitch is simple: one tool for much more than engineering.
🔸 Computer use on your own machine, at least on Mac for now
🔸 Built-in browser
🔸 Image generation
🔸 Deep integrations
🔸 Artifacts inside the app, including tables
🔸 Native automations
This looks like OpenAI’s answer to Claude Code Epitaxy.
📊 @tech