All news
11850
Micro1 Hires People Worldwide to Train Humanoids as the AI Market Demands New Benchmarks
Thousands of workers are filming household chores to train humanoids, while researchers warn that current AI tests say little about how systems will perform on the…

Marcin Moskała reviews GeminiAI: what the audit revealed about coroutines and Android architecture
The developer of the open-source GeminiAI client explains how Marcin Moskała audited the project and why coroutine handling became the real test of its Android…

Coface: AI Threatens Britain's High Salaries and Tax Revenue
Coface ranks the UK as the most vulnerable of 12 developed economies to AI task automation, with high-paid roles and the tax base tied to them at risk.

Saiga Llama 3 8B on 10 GB of VRAM: How Habr Reached 93% Accuracy on War and Peace

Meta prepares to unwind $2 billion Manus acquisition after China blocks the deal

4 non-technical founder patterns that will keep your startup from taking off

Patronus raises €11 million to turn emergency watches for seniors into an everyday device

DeepSeek meets Gemma: a hybrid LLM experiment on Kaggle that broke the Transformers library
Hobbyists grafted four 31B Gemma layers into DeepSeek's MoE architecture without retraining, got around PyTorch and Transformers limits, and booted a chaotic but…

Google Gemma 4 and Qwen 3.6 top the list of best local models for home use in 2026
A 2026 roundup of local models shows an RTX 3060 is now enough for home AI. Pick by VRAM, quantization and task, from coding to transcription.

Google unveils TurboQuant, a 3-bit KV cache for LLMs, but the memory market panicked too early

Rutube moves from a Whisper pilot to its own subtitle platform and speech recognition

Raft shows companies how to evaluate AI agents before putting them into workflows

OpenAI rewrites its founding principles and locks in a commercial course

Sber launches Kandinsky 6 Image: the image generator gets faster and learns to restore photos
Sber has built Kandinsky 6 Image into GigaChat: the model is twice as fast, handles complex prompts better, and can generate, restore and precisely edit photos.

Lightelligence soars 400% after IPO on a bet on optical interconnect for AI
Lightelligence nearly quadrupled on its first day of trading in Hong Kong, as investors bet AI infrastructure's next bottleneck won't be chips but the links between…









