Google's New Model Tops Ranking

Sign Up | Advertise

The AI World Today

• Gemini-exp-1206 Tops Rankings

• E11 Bio Achieves Cost Effective Brain Mapping

• Doubao Tops China’s AI Chatbots

  +

• Heads Up

• AI Solution

Gemini-exp-1206 Dominates Across Chatbot Arena

Google has launched its new AI model, Gemini-exp-1206, now available for free in Google AI Studio, marking a year of exceptional progress for the Gemini series. Dominating the Chatbot Arena rankings, Gemini-exp-1206 has claimed the #1 spot across key categories, including overall performance, hard prompts, coding, math, and instruction following, while also excelling in style control tasks. Key highlights of its improvements since Gemini-exp-1121 include: Securing first place overall (from #2 to #1). Tying with GPT-4o-1120 in style control (from #4 to #1). Matching O1 for the top coding model (from #3 to #1). And taking the lead on hard prompts (from #2 to #1).

E11 Bio Unlocks Affordable Whole-Brain Mapping

E11 Bio has announced a major advancement in brain mapping, achieving a 100-fold reduction in costs and bringing whole-brain connectomics at mouse and human scale within reach. The organization’s PRISM technology assigns neurons unique protein barcodes, enabling them to "self-annotate." This innovation significantly reduces the manual proofreading required to correct errors in AI-based connectome reconstruction, which currently accounts for 95% of costs. Using expansion microscopy, protein staining, and advanced AI, PRISM automates error correction, cutting costs and accelerating progress. E11 Bio predicts achieving a full optical mouse connectome within five years. As a nonprofit committed to open science, E11 seeks collaborators in imaging, automation, and AI to unlock applications in biomedicine and artificial intelligence.

ByteDance’s Doubao Becomes Most Popular Chatbot

ByteDance, the owner of TikTok, is emerging as a frontrunner in China’s generative AI race, investing billions in AI infrastructure and poaching top talent from rivals like Alibaba and Zhipu. The company has become Nvidia’s largest customer in China, acquiring cutting-edge GPUs for AI model development. ByteDance’s AI chatbot Doubao, launched in August 2023, has rapidly become China’s most popular AI application, boasting 60 million monthly active users by November. Founder Zhang Yiming is driving efforts toward artificial general intelligence (AGI), overseeing global AI development teams and launching Cici AI, an overseas chatbot. ByteDance is also developing proprietary AI chips to reduce reliance on Nvidia, aiming to establish a new growth engine as TikTok faces slowing growth and regulatory challenges.

Heads Up 

Aurora, a photorealistic image generator briefly available to Grok users, was replaced by Flux after Elon Musk confirmed it was an internal beta system.

OpenAI's Sora v2, featuring advanced text-to-video, text+image-to-video, and text+video-to-video capabilities with 1-minute outputs, is set to launch imminently.

Meta's Llama 3.3, a 70B open-text model, offers faster, cheaper performance, surpassing GPT-4o and others, with a 128k token context window at $0.10/input token.

Cover completed its first concealed weapon detection system this summer and is now designing a radically improved second-generation system with enhanced capabilities.

Jack Ma, Alibaba founder, highlighted AI opportunities for Ant Group's future during a rare public appearance at the fintech giant's 20th-anniversary celebration on Sunday.

AI Training

Getting Started with Act-One on Gen-3 Alpha

Act-One on Gen-3 Alpha enables users to create expressive animations by combining a driving video with a reference character image. Begin by selecting Gen-3 Alpha in the Generative Video section and clicking the "Act-One" icon on the left sidebar. Upload your source video, ensuring the footage captures the subject’s full face with lively expressions, minimal body movement, and clear audio. While Act-One supports characters at various angles, avoid extreme positions for best results. Next, add a reference image, either by uploading your own or selecting from the provided presets. When all inputs are ready, click "Generate" to produce a seamless animation. Act-One transfers subtle expressions and movements from the source footage to the reference character with precision.

AI Solution

Google’s GenCast Outperforms Leading Weather Model

Google’s DeepMind team has introduced GenCast, an AI-driven weather prediction model that surpasses the European Centre for Medium-Range Weather Forecasts’ ENS, considered the world’s leading forecasting system. According to a study published in Nature, GenCast outperformed ENS in accuracy 97.2% of the time when tested on 2019 forecasts using training data from prior years. Unlike its predecessor, which offered single deterministic forecasts, GenCast generates an ensemble of over 50 predictions, creating complex probability distributions for future weather scenarios. Google is integrating GenCast into Google Search and Maps while also planning to release real-time and historical forecasts for public use in research and modeling. The breakthrough aims to improve accuracy and accessibility in weather prediction globally.