Zhemal Khamidun
Head of AI at Alpina Digital, CPO of AlpinaGPT (~30-person engineering & AI team). Founder of Vibe Academy & Hamidun Consulting, author of the AI channel «Готовим ИИшницу» (@jemal_hamidun).
MetaNCA teaches neural networks to self-organize without backpropagation
The new MetaNCA framework generates weights for neural networks of different architectures through local cellular automaton rules — without classical…
Jet-Long extends language model context windows to 128K without fine-tuning
The Dynamic Bifocal RoPE method allows open-weight models to process up to 128K tokens without retraining, outperforming existing zero-shot baselines and speeding…
Eternis-Forecaster 8B: activation probes are more accurate than reasoning chains and save up to 47% of tokens
Researchers showed that probes of intermediate LLM activations reveal actual forecasts more accurately than chain-of-thought and can save 30–47% of tokens without…
Station F opens new intake for AI-accelerator F/ai for European startups
Why Open AI's Growth Doesn't (Yet) Threaten Anthropic: A Lifecycle Analysis
Meta Launches Muse Image, First Superintelligence Labs AI Photo Model for Instagram
OpenAI Chief Futurist Joshua Achiam Leaves Company After Nine Years in AI Safety
Meta launched Muse — an AI image generator for ads, interior design, and content creators
Meta officially released Muse — a generative image model focused on advertising campaigns, interior design visualization, and content creator tools within Facebook…
Meta allowed Instagram photos to be used for AI generation without explicit user consent
With the launch of the Muse Image model, Meta by default opened access to public Instagram photos — users must opt out themselves.
Liquid AI Released Antidoom: FTPO Method Reduces Hangs in Reasoning Models
French startup ZML released free tool to accelerate AI inference on any chips
Microsoft Replaces OpenAI and Anthropic Models with Its Own AI in Several Copilot Functions
Former DeepMind Director: US Nationalist AI Policy Leads to Catastrophe
NVIDIA Nemotron 3 Ultra sets an accuracy record among open models in LangChain test
LangChain adapted the Deep Agents framework for NVIDIA Nemotron 3 Ultra — the open model delivered the highest accuracy among all open counterparts at a lower cost…
Alibaba researchers discover a DoS vulnerability in reasoning AI models
Scientists from Zhejiang University and Alibaba showed at ICML 2026 that logically contradictory prompts send GPT-o3, DeepSeek-R1, and Gemini 2.5 Flash into useless…