Alexandre Moufarek
@amoufarek
AI R&D Data Strategy @GoogleDeepMind | Gemini Models + AI agents Project Astra & SIMA | ex-Ubisoft (Ghost Recon, Watch Dogs) & ex-Softbank Robotics (Pepper)
ID: 185627004
01-09-2010 13:05:36
3,3K Tweet
762 Followers
288 Following
MarioVGG by Virtuals Protocol is a text-to-video diffusion model that can continuously generate scenes and levels of Super Mario Bros., simulate the physics and movements of a controllable player all through video. Project: virtual-protocol.github.io/mario-videogamβ¦ Paper: virtual-protocol.github.io/mario-videogamβ¦
PALO (Vivek Myers et al., 2024) is a novel approach for few-shot adaptation to unseen tasks that exploits the semantic understanding of task decomposition provided by VLMs. It achieves the same performance as finetuning with >15x less data. Project: palo-website.github.io
WonderWorld (Haoyi Duan et al., 2024) generates 3D scenes interactively. It takes an input image and follows player movement to create a 3D scene in less than 10 seconds. Project: kovenyu.com/WonderWorld/ Paper: arxiv.org/pdf/2406.09394
Imagen 3 & Veo, our best image and video generation models, are coming to Dream Screen to empower YouTube creators. Each creation will be watermarked using SynthID, and YouTube will apply a label that communicates to viewers that it was generated with AI. deepmind.google/discover/blog/β¦
Gemini 1.5 Flash & Pro models got better and cheaper! π²-50% token cost on Pro (both input & output for prompts <128K) π 2x higher rate limits on Flash, ~3x on Pro β© 2x faster output, 3x lower latency πͺ Improved quality in math, long context and vision developers.googleblog.com/en/updated-proβ¦