AI Post — Artificial Intelligence

🗣️ Sam Altman said OpenAI are trying to figure out robotics: “If you could pick one thing to make the US competitive at manufacturing, you would say we need robots that can build a lot more robots” @aipost🏴

🔥Korean robotics leader ROBOTIS has unveiled the 'AI Sapiens' humanoid, a direct challenge to Unitree’s G1 and R1 models. - 4'3" (130 cm) tall and weighing 34 kg. - Features 23 DOF, powered by the brand-new Dynamixel-Q actuators. The robot is a fully open-source hardware system designed to serve as a reliable hardware platform for advanced Physical AI research. AI Sapiens is slated for an H1 2026 launch, with a price point expected in the mid-to-low 10 million KRW range (~$7,000 – $8,700 USD). @aipost🏴

🔥Agents just had their iPhone moment (literally). This swiss AI lab just shipped AI agents that are hooked up to real phones. They don’t use APIs. They don’t use SDKs. They don’t need integrations. Just pixels + touch. Just like humans. Source. @aipost🏴

Unitree G1 on inline skates! 🔥 @aipost🏴

🏓Sony AI built the first robot to beat elite humans in a real sport Sony AI published a Nature paper showcasing Ace, an autonomous table tennis robot that marks a major robotics milestone: the first AI system to defeat elite human athletes in a competitive physical sport under official rules. Why Ace is different: • Ace competes in the messy real world: speed, spin, timing, physics, and pressure. • It reacts, strikes, and adapts in live rallies against trained humans. Insane perception system: • 9 cameras track the ball in 3D space 200 times per second. • Separate gaze systems zoom in mid-flight to detect the ball’s spin axis. • Reads an unmodified ball traveling 20–30 m/s while spinning over 160 rotations/sec. • Full cycle from seeing → deciding → swinging takes 20.2 milliseconds. That’s roughly 10x faster than elite human reaction time. It taught itself, no human coach, no pre-programmed playbook. Trained in simulation for 3,000 hours, playing against itself. Rewarded for stronger shots, more spin, and smarter placement. Results that changed robotics: • In published matches, Ace beat 3 of 5 elite players. • Later, it beat professional players too. • By March 2026, it reportedly defeated multiple pros including Kihara Miyu, ranked around world No. 25. @aipost🏴

🤖OpenAI just gave Codex a second AI boss to approve risky coding actions OpenAI introduced Auto-review, a new autonomous mode for Codex aimed at reducing the constant need for human approvals during long-running software tasks. Instead of stopping every few minutes to ask a developer for permission, Codex can now hand risky decisions to a second independent AI agent that reviews actions in real time. How it works: • Primary Codex agent handles the coding task. • Secondary supervisor agent monitors higher-risk actions before execution. • It can approve steps like: ~ Running tests ~ Launching builds ~ Executing sensitive commands ~ Making workflow-critical changes @aipost🏴

🇨🇳DeepSeek just open-sourced a 1.6T AI model built for coding agents and million-token memory DeepSeek unveiled its V4 generation with two new foundation models: DeepSeek-V4-Pro (1.6 trillion parameters, activating only 49B per token) and DeepSeek-V4-Flash (284B parameters focused on efficiency). The release pushes open-source AI closer to the frontier dominated by closed labs. What makes V4 different: • Introduces DeepSeek Sparse Attention (DSA), a new architecture designed to cut memory and compute costs. • Uses token-wise compression, allowing far longer context windows without the usual infrastructure burden. • Result: 1 million token context becomes standard across official DeepSeek services. Built for coding and reasoning: • V4-Pro is positioned as a new open-source leader in agentic coding and advanced reasoning. • Designed to compete directly with top proprietary models from OpenAI, Anthropic, and Google. • Supports Thinking mode for deeper step-by-step reasoning and Non-Thinking mode for faster responses. DeepSeek’s strategy is clear: don’t just match frontier labs, make frontier AI cheaper, open, and easier to deploy. Source. @aipost🏴

⚠️Anthropic said Mythos was too dangerous to release. Then four random guys in a Discord gained access on day one by guessing the URL... This is pretty insane: → Group in a private Discord guessed the endpoint from Anthropic's naming conventions → They figured out the conventions from the leak in the Mercor breach three weeks ago → Used a contractor's legit eval credentials to walk in → Have been using it ever since to build simple websites The AI that finds zero-days in every operating system on earth was defeated by address bar autocomplete... big yikes 😬 @aipost🏴

📈OpenAI's pre-IPO valuation has officially hit a record $1 trillion. Pre-IPO instruments trading onchain, backed 1:1 by SPV exposure on Jupiter, are providing a real-time proxy for the company’s implied IPO valuation. OpenAI’s implied valuation is now up +163% since October 2025, when rumors first emerged of a potential $1+ trillion IPO. This comes as Anthropic is also nearing a potential $1+ trillion IPO and SpaceX is reportedly targeting $1.7+ trillion. The world has never had this many trillion-dollar private companies. AI is accelerating. @aipost🏴

🦾MIT researchers built artificial muscles powered by AI-controlled fibers MIT researchers have developed robotic fibers that mimic how real human muscles move, contracting, relaxing, and working in bundles just like biceps and triceps. Instead of motors or gears, each fiber contains: • A sealed tube filled with electrically charged liquid • A tiny electric pump • Silent motion driven by fluid flow How it works: • The pump injects electrical charge into the liquid • Ions inside the fluid pull the liquid along • One side contracts while the other expands • Result: smooth, muscle-like movement with no noisy machinery Because they’re fibers, many can be grouped together to increase strength, the same way muscle tissue scales force. In demonstrations, the artificial muscles could: • Bend a robotic arm • Curl a dumbbell • Gently shake a person’s hand This could reshape prosthetics, wearable exoskeletons, humanoid robots, and factory machines by replacing rigid motors with soft, lifelike movement. @aipost🏴

In 2yrs, Google has gone from 0% code written by AI to 75% code written by AI. What an incredible transformation of how software is created. @aipost🏴

⚠️AI becomes the boss: Inside the store run entirely by an algorithm A startup called Andon Labs just handed an AI full control of a physical retail store in San Francisco and told it to make money. The result: a real-world experiment where an AI named Luna…