If a video generation model is allowed to use a non-trivial amount of inference-time compute, how much can it improve generation quality for challenging text prompts?
This is exactly what a new paper from Tsinghua University explores.
GENERATIVE AI
-
Video Generation Models Leverage Inference Compute for Quality
By
–
-
Google Astra: Gemini Live Gets Real-Time Visual Understanding
By
–
Google's Project Astra features started rolling out to Gemini Live
— Rowan Cheung (@rowancheung) 25 mars 2025
The update will allow Gemini to “see” your screen or through your smartphone camera and answer questions real-timepic.twitter.com/KzJ26AP0dKGoogle's Project Astra features started rolling out to Gemini Live The update will allow Gemini to “see” your screen or through your smartphone camera and answer questions real-time
-
Alibaba Qwen2.5-VL-32B-Instruct: Advanced Vision-Language Model
By
–
The Qwen team of Alibaba open-sourced Qwen2.5-VL-32B-Instruct, a new vision-language model —Features enhanced mathematical reasoning and visual capabilities
—Beats Qwen2-VL-72B, Mistral Small 3.1-24B, and GPT-4o-0513
—Apache 2.0 licensed -
DeepSeek V3-0324: Open-Source AI Model with Enhanced Math Capabilities
By
–
DeepSeek released V3-0324, an updated version of its V3 non-reasoning AI
— Rowan Cheung (@rowancheung) 25 mars 2025
—A 641GB model but capable of running on high-end PCs
—Uses MoE, activating only 37B params/token and reducing compute
—Upgraded Math and Coding capabilities
—Open-source under MITpic.twitter.com/sskVI1tIFKDeepSeek released V3-0324, an updated version of its V3 non-reasoning AI —A 641GB model but capable of running on high-end PCs
—Uses MoE, activating only 37B params/token and reducing compute
—Upgraded Math and Coding capabilities
—Open-source under MIT -
AI Startup Reve Dethrones Image Generation Giants
By
–
TODAY'S AI NEWS: A new AI startup, Reve, just dethroned leading image generation giants! Plus, more news from DeepSeek, Qwen, Google, Alibaba, and more. Here's everything you need to know:
-
Reve Image 1.0 Emerges as Top Image Generation Model
By
–
Reve emerged from stealth with a new image-generation model: Reve Image 1.0
— Rowan Cheung (@rowancheung) 25 mars 2025
—Delivers exceptional prompt accuracy, text, and image quality
—#1 in Image Arena, surpassing Imagen 3 and Midjourney v6.1
—Currently free to try with text-based editingpic.twitter.com/qlvgtxAASRReve emerged from stealth with a new image-generation model: Reve Image 1.0 —Delivers exceptional prompt accuracy, text, and image quality
—#1 in Image Arena, surpassing Imagen 3 and Midjourney v6.1
—Currently free to try with text-based editing -
ARC-AGI-2 Benchmark Shows AI Reasoning at 4% Accuracy
By
–
On the newly launched ARC-AGI-2 benchmark, the best reasoning AI today can only achieve 4% accuracy. https://
arcprize.org/leaderboard -
Medium-Sized Questions in Neuroscience and AI
By
–
On medium-sized questions in neuroscience https://
buff.ly/ezlHepe
#AI #MachineLearning #DeepLearning #LLMs #DataScience -
Qwen2.5-VL-32B-Instruct Released with RL Optimizations
By
–
Qwen2.5-VL-32B-Instruct is here! Optimized with RL, it shows significant improvements in human preference and mathematical reasoning.
-
Possible return of GPT 4.5 after beta
By
–
Also, the GPT 4.5 option was gone after the beta phase, but it may return later as a regular option it seems.