Topic: AI Model Efficiency
-
DeepSeek on track for $45B valuation after first funding round
DeepSeek is in talks for its first-ever venture capital funding, with its valuation surging from $20 billion to $45 billion in weeks. The Chinese AI lab gained global attention in early 2025 for training a competitive, open-weight large language model at a fraction of the cost and computing power...
Read More » -
SpaceXAI launches Grok 4.5, an ‘Opus-class model’ per Elon
SpaceXAI has unveiled Grok 4.5, its first major model since going public, designed to handle a full spectrum of tasks including coding, research, and writing. The company claims Grok 4.5 offers twice the token efficiency of competing models, providing significant cost savings at $2 per million in...
Read More » -
Google developing new AI chip to boost Gemini efficiency
Alphabet is developing a new server chip, codenamed "Frozen v2" and expected in 2028, to run its Gemini models with six to ten times greater efficiency in tokens generated per unit of power. The AI industry is increasingly moving toward custom silicon to cut costs and reduce reliance on Nvidia, w...
Read More » -
Mistral's Fast AI Translation Challenges Tech Giants
Mistral AI has released two new open-source speech-to-text models, Voxtral Mini Transcribe V2 and Voxtral Realtime, which support 13 languages and are designed for efficient local processing on devices, enhancing privacy and challenging U.S. tech giants. The company's strategy focuses on creating...
Read More » -
Google's TurboQuant AI Cuts LLM Memory Use by 6x
The high memory demands of large language models (LLMs) are a key factor in current high memory prices, driven by the substantial memory consumption of their key-value caches. Google's new TurboQuant compression technique dramatically shrinks an LLM's memory footprint and accelerates performance ...
Read More » -
How to Make Artificial Intelligence Sustainable
Sasha Luccioni, an AI sustainability researcher, has launched the Sustainable AI Group to help companies reduce the environmental impact of AI, focusing on energy efficiency and the specific demands of tools like speech-to-text translation. Despite the Trump administration's rollback of environme...
Read More » -
Meta to Use Employee Keystroke Data for AI Training
Meta is using internal data from its employees, including keystrokes and mouse movements, to train its AI models for more intuitive digital assistants. The move reflects a broader industry trend of seeking proprietary data sources, as public datasets face copyright and quality issues. This strate...
Read More » -
The Open-Source Rebellion: How Hugging Face is Rewriting Big Tech's AI Playbook
In the high-stakes poker game of artificial intelligence, where tech giants bet billions on proprietary models hidden behind elaborate paywalls, one company is playing an entirely different game. Hugging Face,…
Read More » -
Thrift-maxxing threatens OpenAI and Anthropic IPO plans
Chinese AI models offer dramatically lower prices than American counterparts, with DeepSeek's V4-Pro costing 87 cents for the same output that costs $50 from Anthropic's Fable, driving corporate America toward "thrift-maxxing" by using cheaper models for simpler tasks. Companies like Telnyx, Coin...
Read More » -
Google Launches New TPUs for AI Agents
Google has launched its eighth-generation Tensor Processing Units (TPUs), custom chips designed for what it calls the "agentic era" of AI, moving beyond a simple performance upgrade. The new TPU 8t is specialized for training frontier AI models, while the TPU 8i is dedicated to inference, aiming ...
Read More » -
Google Restructures Chrome Team as OpenClaw Gains Popularity
Google is restructuring its Project Mariner team, signaling a strategic shift away from standalone browser-based AI agents toward more versatile systems that integrate with computer operating systems. Browser automation agents have seen low user adoption and failed to meet commercial expectations...
Read More » -
Anthropic Alleges Chinese AI Firms Used Claude for Training
Anthropic accuses three major Chinese AI firms (DeepSeek, MiniMax, and Moonshot) of creating tens of thousands of fraudulent accounts to conduct millions of unauthorized interactions with its Claude AI system for illicit model distillation. The company warns this unauthorized activity bypasses cr...
Read More »