BentoML unveils llm-optimizer, an open-source tool enabling systematic benchmarking, automated parameter sweeps, and visualization of performance tradeoffs for optimizing self-hosted large language models, replacing ad-hoc trial and error with a constraint-driven, benchmark-focused approach.
In the fierce AI talent war, Chinese tech giant Tencent poaches top AI researcher Yao Shunyu, an expert in AI agents and language models, from OpenAI, luring him with a lucrative compensation package.
Alibaba unveils Qwen-3-Max-Preview, a groundbreaking trillion-parameter language model, boosting AI capabilities with enhanced Chinese-English comprehension, instruction following, and complex query handling, available on Alibaba Cloud and OpenRouter at a premium price.
Powerful AI language models demonstrate remarkable abilities, including reasoning and coding, yet still lack the deep language comprehension of humans, relying on statistical pattern matching rather than true understanding, while smaller models sometimes outperform larger ones on specific tasks.
A quirk in floating-point arithmetic and batch variability in large language models (LLMs) can lead to inconsistent outputs, but achieving deterministic inference through batch-invariant kernels for key operations like RMSNorm, matrix multiplication, and attention resolves these issues.
President Trump and top tech CEOs join forces, announcing massive US investments to power American supremacy in the rapidly advancing artificial intelligence frontier, hailing pro-business policies as fueling AI innovation and growth.
OpenAI collaborates with Broadcom on custom AI chips scheduled for launch in 2024, aiming to meet soaring computing demands and reduce reliance on Nvidia while utilizing the chips internally rather than selling them to customers.
OpenAI projects a staggering $115 billion cash burn by 2029, $80 billion higher than earlier estimates, as soaring costs drive the AI giant to develop its own chips and data centers to power its cutting-edge technology.
Groundbreaking framework Memento unlocks continuous learning for large language models, achieving top performance through a structured memory system that stores experiences and enables case-based reasoning without model fine-tuning, revolutionizing AI's ability to learn from the real world.
Mistral, a French AI startup rivaling OpenAI, secures a staggering €2 billion at a $14 billion valuation, fueling the European AI boom with its open-source language models and chatbot Le Chat, as European AI startups gain momentum with a 55% surge in Q1 2025 investments.