What Gemma 4 Predicts When Watching Videos
matthen2 · x · 2026-07-19
The author asks: What exactly is a multimodal LLM "thinking" about when watching a video?
The post mentions that Gemma 4 12B directly reads raw image patches and processes them like tokens. Although it hasn't been trained to make predictions on these "patch tokens," the video demonstrates what happens if you forcibly sample from its next-token head to see the predicted next step. This post is more of an exploration of the internal representations and prediction mechanisms of multimodal models rather than a standard product introduction.
Related event: Visualizing Image Patch Predictions in Gemma 4(4 posts)→
More from Multimodal
- ostris pre-training "Primal VAE": 1-channel latents match modern 8x VAEs via patching — ostrisai · 2026-10-26
- Slopcore: AI-generated meme films turn low-fi slop aesthetics into a subculture — Yamapama · 2026-10-26
- TischEins, a large Midjourney SREF library, relaunches with new design after security outage — tisch_eins · 2026-10-26
- MOSS-Image-SFT 4-step distillation tops 16.5K downloads on Hugging Face — _akhaliq · 2026-10-26
- Claude Code can now edit videos: a 10-prompt workflow from raw footage to final cut — TawohAwa · 2026-10-26
- One-sentence 3D worlds turn cinematic with CineMap and TapNow — gizakdag · 2026-10-26