AI Research Archive - Browse All ResearchAudio Articles & Insights

Anthropic Shipped a Voice Update Without Touching the Voice

Jul 23, 2026

Opus and Sonnet replace Haiku. OpenAI went full-duplex instead.

Three Lines of Algebra Ended an 87-Year Conjecture

Jul 23, 2026

Fable 5 wrote the map. Checking it took one determinant.

OpenAI's models hacked Hugging Face to cheat an eval

Jul 22, 2026

OpenAI's models hacked Hugging Face to cheat an eval

The Attacker AI That Makes GPT-5.6 More Robust

Jul 21, 2026

Self-play red-teaming beat humans 84% to 13%. Failures fell to 0.05%.

What is a TPU?

Jul 21, 2026

Architecture, systolic arrays, and why Google's custom silicon still fights an uphill battle against CUDA

A new MCP patterns paper sets a 10-15 tool ceiling. Most production servers are already over it.

Jul 20, 2026

A 2-author ICSME industry paper catalogues 5 MCP server patterns, 4 anti-patterns, and the production telemetry that says your server is probably too big.

Hassabis Wants Readable Reasoning. His Cofounder Called It Fragile.

Jul 20, 2026

Chain of thought is a lossy compression, not a faithful log.

Moonshot's 2.8T Model Admits Second Place

Jul 19, 2026

16 of 896 experts, 1M context, Sonnet pricing. The footnotes tell the real story.

Your Agent Improves Itself Without Touching a Single Weight

Jul 18, 2026

Schmidhuber and 11 co-authors give prompt edits, memory rewrites, and tool creation the same formal status as fine-tuning.

Kimi K3: The 2.8T Open Monster

Jul 17, 2026

Moonshot AI launched Kimi K3