OpenAI is enhancing data privacy for its most capable AI models, offering an opt-in zero data retention policy for eligible enterprise clients.

Gemini Live transforms mobile research by allowing users to start deep dives via voice and multitask while AI works.

Submitting to ICLR with numbered citations instead of the required Author-Year format could lead to desk rejection.

Linear attention models, vital for processing massive sequences like DNA, fail critical long-range recall tests, prompting a search for memory and hybrid architectures.

Persistent tool calls, context window limits, and prompt injection cost developers time and money.

InfraAI aims to solve the perennial problem of stale architecture diagrams and fragmented documentation with an AI-augmented, web-based platform.

New GGUF format enables faster inference and reduced memory usage for large language models.

As AI capabilities surge, many outside tech grapple with growing unease about its societal impact, particularly concerning jobs.

An analysis reveals how specific test conditions and outdated benchmarks can artificially boost the perceived effectiveness of sparse attention and KV caching techniques.

A new trie-based algorithm, SALT, drastically cuts chatbot input costs by intelligently reducing token usage.