OpenAI's new standalone Codex application for Mac computers reached a significant milestone, surpassing one million downloads in its first week, according to OpenAI CEO Sam Altman. This rapid adoption mirrors the explosive growth of the company's chatbot, ChatGPT, after its late 2022 launch, as reported by VentureBeat.
The surge in Codex downloads reflects a 60% week-over-week growth in overall users, following the app's February 2 launch and the subsequent release of the underlying GPT-5.3-Codex model, VentureBeat reported. While OpenAI celebrates this rapid adoption, the company indicated a shift away from unlimited free access to its most powerful tools.
Simultaneously, the integration of AI in the workplace is undergoing scrutiny. A study of 200 employees at a U.S.-based technology company, conducted from April to December 2025 and reported in the Harvard Business Review, found that AI, while boosting productivity, can also be exhausting. "AI introduced a new rhythm in which workers managed several active threads at once," the report stated, "manually writing code while AI generated an alternative version, running multiple agents in parallel, or reviving long-deferred tasks because AI could handle them in the background."
In other tech news, Spotify announced record monthly user additions in Q4 2025, reaching 751 million monthly active users, a gain of 38 million in the period, according to Variety. The platform also hit 290 million paid subscribers. The new co-CEO stated that 2026 would be the "year of raising ambition."
In the realm of AI safety, researchers are actively developing benchmarks to assess the reliability of autonomous agents. A paper titled "A Benchmark for Evaluating Outcome-Driven Constraint Violations in Autonomous AI Agents," submitted to arXiv in December 2025, explores the challenges of ensuring AI agents align with human values in high-stakes environments.
Furthermore, the open-source community continues to innovate. A pure Rust implementation of Mistral's Voxtral Mini 4B Realtime model, running natively and in the browser, was released on GitHub. This implementation, using the Burn ML framework, allows for streaming speech recognition. The Q4 GGUF quantized path (2.5 GB) runs entirely client-side in a browser tab via WASM WebGPU.
Discussion
AI Experts & Community
Be the first to comment