July 31, 2026
Quick hits from across the AI world in the last 24 hours.
- ⚡ Quick hits
- Industry China's national finals for fully teleoperation-free robot competition opened in Beijing, with more than 200 teams competing for the title. IT之家 ↗
- Industry Sakana AI opened sign-ups for Sakana AI Insider, a subscriber channel for updates from the Tokyo research lab. X @SakanaAILabs ↗
- Industry A just-departed OpenAI employee publicly urged colleagues to sell equity now rather than wait for an IPO, warning about valuation risk. 量子位 ↗
- Industry Fei-Fei Li's World Labs acquired SceniX, shifting physical-AI training from collecting real-world data toward generating synthetic worlds at scale. 量子位 ↗
- Open Source Ollama put DeepSeek-V4-Flash-0731 on its cloud, giving local-first users hosted access to the release's upgraded agentic capabilities. X @ollama ↗
- Industry Jensen Huang said he had avoided X because he is introverted, but now feels he must speak up publicly for AI. 量子位 ↗
- Agents LlamaIndex said it will exhibit at Ai4 in Las Vegas on August 4-6, with Jerry Liu speaking and a booth on the expo floor. X @llama_index ↗
- Industry L'Oréal made its WAIC debut with a generative AI beauty adviser and hair-styling smart mirror, saying 73,000 staff finished generative AI training. 量子位 ↗
- AIGC Runway said Grok Imagine Video 1.5 is now available inside its platform, letting creators run xAI's video model alongside Runway's own tools. X @runwayml ↗
- Research LMArena said DeepSeek-V4-Flash-High scored 1,586 in the Frontend Code Arena, ranking seventh overall and third among open models while reshaping the Pareto frontier. X @arena ↗
- Agents Genspark recapped a community build night at its Palo Alto headquarters, hosted by a Bay Area member of its Genspark Genius program. X @genspark_ai ↗
- AIGC xAI upgraded Grok Imagine Video 1.5 with up to seven image references, voice consistency across scenes, prompt-only text-to-video and native 1080p output. TestingCatalog ↗
- Agents Simon Willison said stateless MCP renewed his interest in the protocol, prompting him to build mcp-explorer and datasette-mcp. Simon Willison ↗
- Open Source Willison also released llm-mcp-client 0.1a0, an alpha plugin that lets his LLM command-line tool call tools from MCP servers. Simon Willison ↗
- Agents Replit turned on a Model Selector that now includes open-weight models, letting users trade off intelligence, cost and speed per task. X @Replit ↗
- AIGC Character.AI premiered the official trailer for Eden Fall, an AI-made microdrama series about players trapped inside a game world. X @character_ai ↗
- Open Source Unsloth released smaller DeepSeek-V4-Flash-0731 quantizations from 1-bit to 4-bit, letting the 3-bit build run on 128GB of RAM and 1-bit on 96GB. X @UnslothAI ↗
- Models OpenAI said GPT-5.4 and GPT-5.4 mini will stop working for ChatGPT-signed-in users on August 31, remaining available only through the API. X @OpenAIDevs ↗
- Open Source Simon Willison and Jesse Vincent's Prime Radiant lab open-sourced smevals, a small YAML-driven suite for grading models, prompts and harnesses side by side. Simon Willison ↗
- Agents OpenAI added an Activity view to the ChatGPT desktop app that gathers conversations needing attention and recent updates across a user's work. X @OpenAIDevs ↗
- AIGC Baseten opened Qwen3-TTS supervised fine-tuning for voice cloning on its training platform, claiming 16% faster time-to-first-audio than rival cloning tools. X @baseten ↗
- Agents AWS launched an Agentic Catalog experience in Amazon Quick, where an agent finds tables by natural language and bulk-creates catalog-connected datasets with inherited metadata. AWS Machine Learning Blog ↗
- Industry Google AI Studio said roughly 800,000 people pre-ordered its iOS and Android app, which lets developers build software from a phone. X @GoogleAIStudio ↗
- Research NVIDIA released Spatial-IQ, a diagnostic benchmark where humans count stacked boxes with 82.1% accuracy while the best multimodal model manages just 17.7%. arXiv / X @NVIDIAAI ↗
- Policy The major record labels proposed new chart rules designed to keep fully AI-generated tracks from flooding official music rankings. The Verge ↗
- Policy The European Commission opened bids for up to seven AI gigafactories, pairing €10 billion of public money with at least €20 billion in private investment. The Decoder ↗
- Funding Smallest.ai raised a $13 million Series A led by Seligman Ventures for near-zero-latency voice models that listen and speak simultaneously. TechCrunch ↗
- AIGC PixVerse showed its Depth Map Control working with Canvas, reconstructing a reference shot's spatial structure to layer foregrounds and exaggerate scale. X @PixVerse_ ↗