Future Science & Tech

AI Model & Agent Releases

• 🔥 GLM-5.1: GLM-5.1 is out on Hugging Face as #1 in open source and #3 globally across SWE-Bench Pro, Terminal-Bench,...

Wave of XR advancements for deeper, context-aware experiences:

• Gemini AI in mixed reality headsets/smart glasses hears and sees what you're...

Unreal Robotics Lab alpha released as an open-source Unreal Engine plugin integrating full MuJoCo physics for photorealistic rendering and accurate contact physics—with no compromises.

Absurd edge ML push: Train a real single-neuron classifier via gradient descent, quantize 3 params (2 weights + bias) into a 1x1 PNG's RGB...

Vanast enables virtual try-on featuring human image animation powered by synthetic triplet supervision. A promising leap for realistic AR shopping experiences.

• Zero-Width Chars: Super high bandwidth, but fragile – stripped by Slack/Twitter/email and trivially flagged by scanners.
• Homoglyphs: Cyrillic...

Emerging papers push RL boundaries for AI agents:

• Cog-DRIFT adapts problems using Zone of Proximal Development—simplifying via reformulation to...

GLM-5.1 surges ahead as the new open-source AI frontrunner for long-horizon tasks:

• Benchmark dominance: #1 in open source, #3 globally on SWE-Bench...

QoderWork revolutionizes workflows by acting on your desktop:

• Opens files, analyzes data, runs code—all locally.
• Automates multi-step tasks like...

New preprint uncovers the 'boiling frog' effect of AI use in cognitive tasks:

• RCTs show: Just 10 minutes of AI assistance leads to worse...

Autonomous discovery era dawns with CORAL's multi-agent systems for open-ended science – but current agents remain framework-bound.

• Stanford...

Consumer Apple tech has transformed from earthly gadgets to orbital tools, starting with Mac Portable on Shuttle STS-41/43 for microgravity HCI tests,...

• Epic scale: 8 bimanual setups, 100+ hours demos, 5k+ GPU hours to teach folding
• Flashy demos everywhere, but real story? Data, failures, raw progress
• Key step toward household humanoids mastering chores

Quick catch-up on March's AI pulse:

• New innovations profiled alongside product drops
• AI's job impact in focus, amid breakthroughs and fights
• Answering What's new in AI right now?

Emerging tensions meet rapid advances—watch jobs and society.

• RLVR boosts LLM reasoning but stalls on hard problems: failed rollouts mean zero learning signal.
• Unsolvable visuals keep training stuck.
• Cog-DRIFT breaks this barrier for deeper reasoning gains.

Practical neurotech breakthrough from Dylan Gaines at Kennesaw State:

• Noninvasive headset reads scalp brain signals as users eye letter groups or...

Urgent quantum risk alert: AI optimized Oratomic's algorithm, slashing atoms per qubit from 100-1,000 to just 3—a 100x reduction.

• Enables smaller...

OpenWorldLib introduces a unified codebase and definition for advanced world models, with a new paper detailing the framework. A key step toward standardizing AI world simulations.

Researchers fine-tuned a video model solely on purely synthetic data for 3D scene relighting, yet it generalizes to real captured scenes—handling shadows, reflections, and all. Breakthrough for dynamic lighting in AR/VR.

PyTorch-forked hybrid attention delivers 51x faster inference on consumer GPUs.

• Full attention: O(n²), 17.96s / 5.6 tok/s
• Hybrid: O(n·W + n·D),...