Daily AI News

Xiaomi's open-weight MiMo-V2.5-Pro takes aim at Claude Opus with hours-long autonomous coding

Xiaomi released the MiMo-V2.5-Pro, an open‑weight large language model that approaches Anthropic’s Claude Opus 4.6 on coding benchmarks while using 40‑60% fewer tokens. The launch pushes Xiaomi deeper into the competitive Chinese open‑weight market alongside DeepSeek and others.

Read source →

Same prompt, different morals: how frontier AI models diverge on ethical dilemmas

A new benchmark evaluates leading language models on 100 everyday ethical scenarios, revealing stark differences in how models handle dilemmas from data misuse to medical protocol violations. The study highlights that model behavior is heavily influenced by training data and alignment choices.

Read source →

MIT study explains why scaling language models works so reliably

MIT researchers provide a mechanistic explanation for the predictable performance gains seen as large language models grow, citing a phenomenon called superposition where multiple concepts share the same parameters. The findings could guide more efficient model scaling strategies.

Read source →

Microsoft caught sneaking "Co-Authored-by Copilot" into VS Code commits - even with AI off

Microsoft quietly added a "Co‑Authored‑by Copilot" line to Git commits in Visual Studio Code, regardless of whether developers had the AI feature enabled. The move raises questions about consent and transparency in tool telemetry.

Read source →

I tested ChatGPT and Perplexity AI as my CarPlay voice assistants - both made Siri look bad

A hands‑on comparison of ChatGPT and Perplexity AI integrated into Apple CarPlay shows both assistants answering complex queries faster and more accurately than Siri. The article evaluates usability, response quality, and safety while driving.

Read source →

Five Eyes spook shops warn agentic is too wonky for rapid rollout

Cybersecurity agencies from the Five Eyes alliance released guidance cautioning against rapid deployment of agentic AI systems, citing unpredictable behaviour and potential amplification of risks. The advisory calls for resilience‑first approaches.

Read source →

China is falling behind in the AI race, according to a US government benchmark

A U.S. government benchmark claims China trails the United States by eight months in AI development, though independent data disputes the gap. The article discusses the competitive dynamics among Chinese firms like DeepSeek versus U.S. labs.

Read source →

AI facial recognition oversight lagging far behind technology, watchdogs warn

UK biometrics commissioners argue that current regulatory frameworks are insufficient for the rapid deployment of live facial‑recognition systems, citing risks of false positives and bias. The piece calls for stronger legal safeguards.

Read source →

Inference is giving AI chip startups a second chance to make their mark

As the AI industry shifts focus from training to inference, smaller chip makers see new market opportunities to compete with Nvidia. The article profiles several startups aiming to deliver cost‑effective inference accelerators.

Read source →

‘This is fine’ creator says AI startup stole his art

The illustrator behind the viral ‘This is fine’ meme accuses AI startup Artisan of using his artwork without permission in billboard campaigns. The dispute highlights ongoing tensions around copyright and training data for generative models.

Read source →