7th January - AI News Daily - xAI Raises $20B, OpenAI Ordered to Release Logs as AI Scrutiny Intensifies

Send a text

🌍 INAI • The Open AI Hub

The Intelligence Atlas → the world’s most comprehensive, open hub of AI knowledge. 2 Million+ tools, models, agents, tutorials & daily news—free for all, updated every day.

https://github.com/inai-sandy/inAI-wiki

Top Highlights: xAI raised $20B and started training Grok 5 with self-powered infrastructure; Arena secured $150M at $1.7B valuation for multimodal evaluation; NVIDIA expanded robotics through Isaac-LeRobot integration and open-sourced Alpamayo; a federal judge ordered OpenAI to release 20M ChatGPT logs; Amazon launched Alexa+ online while Google TV added Gemini controls.

New Tools: OpenSRC CLI automates AI agent deployment; LlamaSheets structures messy spreadsheets; Sharp enables single-image-to-3D on consumer GPUs; DatologyAI DatBench cuts benchmarking costs 10x; Unsloth MLX accelerates Apple on-device training; Lightricks LTX-2 delivers synchronized text-to-video-and-audio.

LLM Updates: GPT-5.2 reportedly solved an Erdős problem; NousResearch released NousCoder-14B with reproducible RL; DFlash decoding achieves 6.2× speedups; Liquid AI LFM2.5 runs on Raspberry Pi; Mistral OCR 3 leads in document accuracy; Korea Telecom launched Mi:dm K 2.5 Pro.

Research: Artificial Analysis Intelligence Index v4.0 introduces new evaluation metrics; Apple showed 32% training time savings via hyperparameter transfer; SGD with batch size 1 trained 1.3B parameter models; study finds only 2.5% of tasks fully automatable.

Industry: NVIDIA-Hugging Face-DeepMind-Boston Dynamics aligned on robotics; OpenAI faces intensified legal scrutiny over data practices; consumer platforms shifted with Alexa+ web launch and ChatGPT leaving WhatsApp; safety concerns grew over AI-aided pathogen instructions and deepfakes.

Tutorials: Free RL masterclass (Jan 15) and Claude Code workshop; scaling document processing patterns; local RAG security recipe; Reachy Mini assistant build; agent iteration playbooks; FinePDFs handbook.

Showcases: Reachy Mini on Raspberry Pi 5 at CES; LTX-2 identity-preserving animation; Kling AI Motion Control experiments; Perplexity multi-model orchestration; LLM poker tournament (20,000+ hands).

Discussions: Smarter learning may beat brute-force scaling; better evals replacing brittle leaderboards; agent design shifting to persistent memory; sim-only training limitations; coding agents vs. artisanal craftsmanship debates; OpenAI breadth vs. Anthropic focus strategies.

Support the show

🌍 INAI • The Open AI Hub

The Intelligence Atlas → the world’s most comprehensive, open hub of AI knowledge. 2 Million+ tools, models, agents, tutorials & daily news—free for all, updated every day.

https://github.com/inai-sandy/inAI-wiki