/ SECTION

Research.

Scientific breakthroughs, papers and academia.

Latest Models Research Policy Tools Industry Open Source Newsroom

I FOKUS

OpenAI Solved an 80-Year-Old Math Problem — and Mathematicians Confirm It

OpenAI's reasoning model has disproven a geometry conjecture from 1946 — and this time, prominent mathematicians stand behind the result.

17 MIN AGO · 4 MINLES ⟶

Underground

Stanford DSL beats Triton 14x — and nobody is talking about it

ThunderKittens from Stanford's Hazy Research Lab crushes established tools in GPU performance — and the community is starting to notice.

1D AGO

Underground

Google crushes GPU memory with TurboQuant — 8x performance on 4-bit

A technical deep dive on Lobsters AI goes viral: TurboQuant from Google Research compresses the KV cache in LLMs to just 3 bits — without sacrificing accuracy.

2D AGO

Models

Gemini Omni transforms images, audio, and text into video in one conversation

Google has launched Gemini Omni Flash – a multimodal AI model that generates and edits video from text, images, and audio through natural conversation. According to Google DeepMind, this is a step towards AGI.

3D AGO

Models

NVIDIA's new AI model beats competitors with 3.3x higher performance

NVIDIA launches Nemotron 3 Nano Omni – a compact, open multimodal AI model with a one-million-token context window that outperforms rivals from Qwen and OpenAI in inference speed.

APR 28

Flere saker.

APR 28ResearchDeepMind Legend Raises $1.1 Billion: Aims to Train AI Without Human Data APR 27UndergroundGoogle's TurboQuant Crushes the KV-Cache Bottleneck — 6x Memory Reduction Without Quality Loss