Character.AI, a platform offering AI chatbots for socializing and role play, has released a video generation model called AvatarFX in closed beta. Promising the ability to make photorealistic images “come to life — speak, sing and emote — all with the click of a button,” the technology combines audio and video to create a variety of visual style and voice, from realistic 3D — including “non-human faces (like a favorite pet)” — to 2D animations, according to the company. AvatarFX also has the ability “to maintain strong temporal consistency with face, hand and body movement” and can “power videos with multiple speakers.” Continue reading Character.AI Introduces New Video Generator in Closed Beta

Google has debuted a new accelerator chip, Ironwood, a tensor processing unit designed specifically for inference — the ability of AI to predict things. Ironwood will power Google Cloud’s AI Hypercomputer, which runs the company’s Gemini models and is gearing up for the next generation of artificial intelligence workloads. Google’s TPUs are similar to the accelerator GPUs sold by Nvidia, but unlike the GPUs they’re designed for AI and geared toward speeding neural network tasks and mathematical operations. Google says when deployed at scale Ironwood is more than 24 times more powerful than the world’s fastest supercomputer. Continue reading Google Ironwood TPU is Made for Inference and ‘Thinking’ AI

OpenAI is getting close to finalizing its first custom chip design, according to an exclusive report from Reuters that emphasizes the Microsoft-backed AI giant’s goal of reducing its dependency on Nvidia chips. The blueprint for the first-generation OpenAI chip could be finalized as soon as the next few months and sent to Taiwan’s TSMC for fabrication, which will take about six months — “unless OpenAI pays substantially more for expedited manufacturing” — according to the report. Even by usual standards, the training-focused chip is already on a fast track to deployment. Continue reading OpenAI In-House Chip Could Be Ready for Testing This Year

Model training continues to hit new lows in terms of cost, a phenomenon known as the commoditization of AI that has rocked Wall Street. An AI reasoning model created for under $50 in cloud compute credits is reportedly performing comparably to established reasoning models such as OpenAI o1 and DeepSeek-R1 on tests of math and coding aptitude. Called s1-32B, it was created by researchers at Stanford and the University of Washington by customizing Alibaba’s Qwen2.5-32B-Instruct, feeding it 1,000 prompts with responses sourced from Google’s new Gemini 2.0 Flash Thinking Experimental reasoning model. Continue reading Reasoning Model Competes with Advanced AI at a Lower Cost

Meta Platforms has introduced an AI model it says can generate 3D images from text prompts in under one minute. The new model, called 3D Gen, is billed as a “state-of-the-art, fast pipeline” for turning text input into high-resolution 3D images quickly. The app also adds textures to AI output or existing images through text prompts, and “supports physically-based rendering (PBR), necessary for 3D asset relighting in real-world applications,” Meta explains, adding that in internal tests, 3D Gen outperforms industry baselines on “prompt fidelity and visual quality” and for speed. Continue reading Meta’s 3D Gen Bridges Gap from AI to Production Workflow

Meta’s next generation AI silicon is a 5nm chip designed to power the models that provide recommendations to those who use its social network platforms. The new MTIA inference accelerator is part of a “broader full-stack development program for custom, domain-specific silicon that addresses our unique workloads and systems,” Meta says. The next-gen MTIA more than doubles the compute and memory bandwidth of its predecessor, the 7nm MTIA v1 chip introduced in May 2023, resulting in 3x the performance, according to Meta, which says the new silicon is already live in 16 data centers. Continue reading Meta Deploys Gen 2 MTIA AI Accelerator Chip in Data Centers