🤖 AI Summary
Elon Musk’s xAI is training video-generation features for its Grok chatbot by having employees annotate short clips — including five- to 10-second snippets from Hollywood films such as Universal’s Hellboy II — as part of internal projects called “Vision” and “Moongazer.” Workers labeled shot composition, camera depth and view, cinematography style, lighting, scene settings and individual objects, and in Moongazer identified transitions, captions and other on-screen elements. Musk has said Grok Imagine will debut "watchable" full-length films by the end of 2026 and "really good movies" in 2027, and Business Insider reporting found xAI actively using high-quality cinematic footage to teach the model film grammar and visual storytelling.
The work highlights a larger industry fault line: training cutting‑edge generative models often relies on copyrighted, studio-grade content, which studios say can be infringing while developers argue it’s necessary for high-quality outputs. Legal experts note potential infringement at every stage — downloading, storing, filtering and generating — and recent cases (e.g., Midjourney suits, Anthropic’s $1.5B settlement) show heavy stakes. Companies are experimenting with guardrails and rightsholder controls, but xAI’s terse reply (“Legacy Media Lies”) and tests that produced Hellboy-like images (“Heckboy”) underscore the unresolved tension between model capability, legal risk, and who ultimately profits from creative training data.
Loading comments...
login to comment
loading comments...
no comments yet