Today in Generative Media

Whisper hallucinates; AI world models; AI slop floods Medium; Perplexity seeks $9B

News and Opinion

• Researchers say an AI-powered transcription tool used in hospitals invents things no one ever said (AP News)

• What are AI ‘world models,’ and why do they matter? (TechCrunch)

• AI Slop Is Flooding Medium (Wired)

• Perplexity AI seeks valuation of about $9 billion in new funding round (CNBC)

• AI Detects Mysterious Detail Hidden in a Famous Raphael Masterpiece (ScienceAlert)

• AI helps humans have a 20-minute 'conversation' with a humpback whale named Twain (Earth.com)

• Nvidia overtakes Apple as world's most valuable company (Reuters)

• We finally have an ‘official’ definition for open source AI (TechCrunch)

• Google Pixel 10 and 11 leak reveals new AI tools and a big camera update (The Verge)

• Apple debuted AI on the iPhone today. Here’s what to look out for (CNN)

• I created a chatbot of myself and had it answer my Instagram DMs. Boy, was I annoying. (Business Insider)

Software

• Meta releases an ‘open’ version of Google’s podcast generator (TechCrunch)

• Flux-based IC-Light Model with 16ch VAE and native high resolution (HuggingFace) Examples in GitHub discussion

• Voice Chat with Anyone, powered by F5-TTS + QWEN (X)

• As bit of a personal challenge I decided to get the new Mochi text2video model running locally on 4090, this was generated under 20GB VRAM in ComfyUI. (X)

Research

• Binocular3DGS: Binocular-Guided 3D Gaussian Splatting with ViewConsistency for Sparse View Synthesis (NeurIPS 2024 project page)

• DiffGS: Functional Gaussian Splatting Diffusion (NeurIPS 2024 project page)

• DAWN: Dynamic Frame Avatar with Non-autoregressive Diffusion Framework for Talking Head Video Generation (project page)

• Inkspire: Sketching Product Designs with AI (UIST 2024 poster, ACM)