So what's the deal with Amazon Nova? They released Nova 2 in December, and even then, the top flight Nova 2 model trailed Sonnet 4.5. And it still hasn't left preview.
RESEARCH
-
Gemini analyzes billboard advertising industries with AI
By
–
I asked Gemini: "Can you analyze all the billboards listed on https://
101ads.org and give me a report of what industry is represented by the company on each billboard" -
Revela: Efficient AI Retrievers Without Expensive Annotated Datasets
By
–
What if we could train powerful AI retrievers to find specialized information without needing huge, expensive datasets?
— 机器之心 JIQIZHIXIN (@jiqizhixin) 9 avril 2026
A team from TU Darmstadt, University of Washington, CMU, Microsoft, and Tencent AI Lab presents Revela!
Revela leverages self-supervised language modeling. It… pic.twitter.com/bC4emjEYdVWhat if we could train powerful AI retrievers to find specialized information without needing huge, expensive datasets? A team from TU Darmstadt, University of Washington, CMU, Microsoft, and Tencent AI Lab presents Revela! Revela leverages self-supervised language modeling. It teaches retrievers to understand semantic relationships between document segments by predicting "next chunks" of info, integrating retriever similarity scores directly into this learning process. Without any annotated query-document pairs, Revela surpasses larger supervised models and proprietary APIs in code and reasoning-intensive domains. It achieves unsupervised SoTA on general benchmarks with ~1000x less data and 10x less compute! Revela: Dense Retriever Learning via Language Modeling Paper: openreview.net/forum?id=e7pA… Code: github.com/TRUMANCFY/Revela Model: huggingface.co/trumancai/Rev… Our report: mp.weixin.qq.com/s/9TmVSNHMQ… 📬 #PapersAccepted by Jiqizhixin
-
Small AI Model Advancement Outpaces Large Model Development
By
–
Our rate of advancement with the small model has been so fast that the large model has not yet caught up. V15 will be the large model.
-
Digital twin of world created with millimeter accuracy
By
–
They used that data to build a digital twin of the world down to 1 mm accuracy. It is sick
-
8 Types of Visualization Techniques Used in Data Analytics
By
–
8 Types of Visualization Technique Used in #DataAnalytics
— Ronald van Loon (@Ronald_vanLoon) 9 avril 2026
by @Python_Dv
#DataScience #BigData pic.twitter.com/L6oiYaR6n08 Types of Visualization Technique Used in #DataAnalytics by @Python_Dv #DataScience #BigData
-
Multimodal Protein Design Enables DNA-Based Chemistry Encoding
By
–
General Multimodal Protein Design Enables DNA-Encoding of Chemistry Rector-Brooks et al.: https://
arxiv.org/abs/2604.05181 #ArtificialIntelligence #DeepLearning #Chemistry -
AI Designs Multimodal Proteins to Encode Chemistry in DNA
By
–
General Multimodal Protein Design Enables DNA-Encoding of Chemistry
— AGI.Eth (@ceobillionaire) 9 avril 2026
Rector-Brooks et al.: https://t.co/dY9Zho50xY#ArtificialIntelligence #DeepLearning #Chemistry pic.twitter.com/iSwMFXQWHwGeneral Multimodal Protein Design Enables DNA-Encoding of Chemistry Rector-Brooks et al.: arxiv.org/abs/2604.05181 #ArtificialIntelligence #DeepLearning #Chemistry
-
Anthropic’s Glasswing Promotion: A Flawed Zero-Knowledge Proof
By
–
A zero-knowledge proof is a way to convince you I know something without actually telling you what I know. Anthropic’s promotion of Glasswing is like a zero-knowledge proof without the proof part: you don’t gain any knowledge, and you may not be convinced, but you have a sense
-
AGIBOT Genie Sim 3.0: Full Stack Embodied AI Infrastructure
By
–
AGIBOT just dropped Genie Sim 3.0 🤖@AGIBOTofficial is turning embodied AI into a full stack: environment → data → training → evaluation, all in one system.
— RoboHub🤖 (@XRoboHub) 9 avril 2026
Text → fully interactive 3D worlds in minutes, with aligned RGB, depth, and LiDAR.
Built-in benchmarks across… pic.twitter.com/f60bgBVmjlAGIBOT just dropped Genie Sim 3.0 🤖 @AGIBOTofficial is turning embodied AI into a full stack: environment → data → training → evaluation, all in one system. Text → fully interactive 3D worlds in minutes, with aligned RGB, depth, and LiDAR. Built-in benchmarks across instruction following, spatial reasoning, manipulation, robustness, and sim-to-real — designed to actually measure what matters in deployment. Deep RL integration (RLinf), 1000Hz physics, massive parallel simulation, and Gym-compatible pipelines. This is the real play: scaling data, standardizing evaluation, and closing the loop between simulation and real-world robots. Embodied AI isn’t just about better models anymore — it’s about infrastructure that can actually scale.