The “Godfather of AI” Geoffrey Hinton and a vocal critic of its potential dangers, has shifted his perspective. He now sees a “ray of hope” in AI. Hinton argues that attempting to “dominate” AI as an assistant will be futile. Instead, he proposes designing AI with a “maternal

Columbia CS Professor explains why large language models (LLMs) can’t generate new scientific ideas. This is because LLMs learn a structured “map” or “Bayesian manifold” of known data and perform well within that framework but struggle outside it. However, true discovery

Can AI invent new mathematics? A new paper from DeepMind and renowned mathematician Terence Tao demonstrates how. Using AlphaEvolve, the team combines LLM-generated ideas with automated evaluation to propose, test, and refine mathematical algorithms. In tests on 67 problems

From [MASK] to meaning. A distilled model's focus and a merged model's power, visualized through diffusion.

Gemini Nano Banana Sundar Pichai

They ship the #1 model on Monday and publishes ‘deep learning is an illusion’ on Tuesday. Peak Google behavior 😂

The scary part is when Gemini says “My intuition ….”

Meet the 7-million-parameter model that just humiliated every frontier giant on hard Sudoku: 87% solved vs GPT-4’s 0%. No 1000x more parameters, no 1000x more compute—just a tiny 2-layer transformer that thinks 21 times about the same puzzle and gets better every loop. Samsung

From PayPal to the Kardashev Scale in 114 minutes — Elon Musk unfiltered with Nikhil Kamath. This isn’t an interview, it’s a masterclass on the future.

EV FLIPPENING IS HERE 🚀 BYD just dropped megawatt chargers that add 400 km in 5 MINUTES. That’s the exact time it takes to fill a gas tank. Game. Over. ⚡️💨

Elon nails the physics: In vacuum, blackbody radiation reigns supreme. Stefan-Boltzmann scaling dumps waste heat at ~P ∝ T⁴ — zero watts on active cooling (Earth supercomputers bleed 40-50% efficiency fighting convection). Pair with constant ~1366 W/m² solar flux (no

Demis Hassabis just dropped the truth bomb: Robotics didn't fail because of hardware... it failed because we lacked TRUE intelligence. But now? Gemini-level models are the 'brain' robots have been waiting for. When the software gets smart, the hardware explodes forward. AGI

At Davos, when the IMF pegged India as a 'second-tier' AI contender, Minister Ashwini Vaishnaw delivered a masterclass in rebuttal: We're not just in the game—we're architecting it across all five layers (applications, models, chips, infra, energy). Stanford's AI Index ranks us

x.com/i/article/2026…

When America sneezes, does India still catch pneumonia? Foreign money still matters hugely in Indian markets — US-based funds alone hold ~44% of all FPI assets. When Wall Street panics, that money sprints home into safe US Treasuries. But here’s what’s changed: FPI ownership has