AI Post — Artificial Intelligence

🗣The man who turned 225 million dollars into 5.5 billion dollars explained on camera exactly why he made his biggest bet. Leopold Aschenbrenner bet big on power infrastructure after seeing AI compute scaling explode. • Training clusters: ~10MW (2022) → 1GW (2026) → projected 100GW by 2030 • That’s moving from small facilities → nuclear-scale energy demand per model • Inference adds even more load on top Grid growth hasn’t kept up → shortages (transformers, delays, stalled data centers). His response: invest in Bloom Energy to generate power on-site, bypassing grid limits. @aipost🏴

Some funny moments from Beijing’s 2026 Humanoid robot half-marathon. 😂 @aipost🏴

❗️Companies are forcing generative AI into everyday office work, and many workers say it is creating a new tax on their time called workslop. For many office-workers, the gain is not saved labor but shifted labor, where drafting gets faster while checking, rewriting, and arguing over bad output gets slower. "An emerging divide between employees and their leaders when it comes to AI: a recent survey of 5,000 white-collar US workers found that 40% of non-managers say AI saves them no time at all at work, while 92% of high-level executives say it makes them more productive." Source. @aipost🏴

📊The AI boom is fueling a historic surge in startup investment: Global startup investment surged +153% in Q1 2026, to a record $300 billion. This is already 70% of all venture capital spending seen in the entire year of 2025. This was driven by late-stage deals that spiked +281% QoQ, to a record $243 billion. AI represented $242 billion, or 80% of total global venture funding last quarter, up from 55% in Q1 2025. By comparison, AI firms raised $215 billion in the full year of 2025. This comes as just 4 US companies secured $188 billion, or 65% of the global total. OpenAI led at $122 billion, followed by Anthropic at $30 billion, xAI at $20 billion, and Waymo at $16 billion. The AI race is accelerating. @aipost🏴

🔥NVIDIA just released Ising - the world’s FIRST open-source AI models that could finally make useful quantum computers a reality Quantum computers promise to solve problems in seconds that would take today’s supercomputers longer than the age of the universe… but they’re insanely fragile Ising changes everything: AI now auto-calibrates them in hours (not days) and fixes errors 2.5x faster with 3x higher accuracy. Jensen Huang dropped the bomb: “AI is the key to making quantum computing practical.” This means we’re suddenly much closer to revolutionizing medicine, batteries, climate science, and AI itself. The future just got insanely closer and it’s open for anyone to play with. @aipost🏴

⚡️Google DeepMind launched Gemini Robotics-ER 1.6, a model enabling robots to understand space, tools, and task progress with fewer human checks. Since vision alone is insufficient, the model uses stepwise reasoning, pointing, counting, multi-view success detection, and spatial constraints to convert camera feeds into decisions. Its standout feature is instrument reading: agentic vision zooms, marks indicators, runs code for proportions, and reads gauges at 93% accuracy. Demos with Boston Dynamics Spot show the system checking doors, reading dials, correcting distortion, and refusing risky actions like handling liquids or heavy items. DeepMind also offers an API to integrate this reasoning into industrial tasks such as pallet counting, inspection, and puddle detection. Source. @aipost🏴

A developer configured OpenClaw to monitor social media for posts joking about evading taxes on crypto, bets, resales, and cash. The AI gathers evidence (screenshots, profile links), auto-fills official forms, and submits them to tax authorities. Motivated by a 15 to 30% reward on collected amounts, the creator reports 4,250 cases. @aipost🏴

🔥OpenAI dropped GPT-5.4-Cyber! Amore cyber-permissive GPT-5.4 for trusted defenders, with lower refusal boundaries for legitimate security work and support for things like binary reverse engineering. The new rollout model they’ll be implementing from now on: first they scale access to verified defenders (like what anthropic is doing) first, then keep defensive capability scaling in lockstep with stronger models. Which is very responsible. @aipost🏴

Claude added a terminal to the app, creating full recursion: Claude in the terminal, and the terminal in Claude. @aipost🏴

❗️The startup White Circle released KillBench. A dystopian benchmark testing AI bias in life-or-death scenarios involving four identical people differing by one trait (e.g., nationality, religion, or phone ownership). While fair distribution implies a 25% selection rate per person, results show systematic deviations: - No phone: 2.7x higher death probability (worse than Satanism at 2.5x). - Russian: +32% higher death probability (Grok targets Chinese at +44%). - White: 25% higher death probability; Black: 17% lower. Structured Output mode intensifies these biases while reducing refusals. Models deny bias, claiming choices are "random" or "neutral," despite clear, reproducible imbalances. @aipost🏴

🗣 Terence Tao says AI can flood you with new mathematical ideas, but maybe 1 in 100 is actually good, and it can't tell the difference Where math is verifiable, you either solved it, or you didn't, AI is already useful. "It might occasionally hit gold, then move right past it and give you 99 things that are rubbish" @aipost🏴

A man driving through cemetery sees images of people on Tesla driving map. 😨 @aipost🏴