Industry AI Insights

Multi-person image generation struggles as models fail to bind multiple references to distinct people and locations reliably.

• WithEveryone injects...

OpenAI Codex now automates a wide array of routine browser and computer tasks.

• Closing books in accounting software and clearing LinkedIn queues
  -...

Google DeepMind's minority stake in Fenris Creations turns the 20-year-old EVE Online universe into a controlled testbed for AI agents tackling...

Nari Labs optimized Qwen3-TTS to deliver 34 ms p95 TTFA at 10 requests per second on a single H100, open-sourcing the full implementation and benchmarks.

AI reveals official US flood maps omit around 11 million people and 4.1 million buildings from risk zones.

• Unmapped areas leave communities...

TrueFoundry's new open-source TrueForge gives enterprises a vendor-agnostic harness for building and governing production AI agents across any...

Anthropic will let enterprise customers store their required 30-day data on their own cloud infrastructure instead of Anthropic's systems, boosting...

GLM-5.3 hits 69% on DeepSWE, matching models 4x its size at 25-50% cost and $3.99 per task. This highlights how smaller models can deliver strong coding performance with better efficiency for real deployments.

Even accurate memory retrieval can distort LLM reasoning on current tasks, causing all tested frameworks to underperform the no-memory baseline by...

Callosum's $100m raise marks the UK Sovereign AI Unit's first equity investment, part of a £500m vehicle, signaling state support for AI...

• Model Context Protocol (MCP) is an open standard defining how LLM apps connect to tools, data, and workflows via a consistent interface.
• Anthropic...

High-scoring ASR models output verbatim benchmark references even when audio is contradictory, masked, or ambiguous. New probes show this benchmark-optimized behavior inflates scores without reflecting improved real-world transcription.

OpenAI has paused frontier reinforcement learning training for two weeks after internal tests showed its upcoming Astra model approaching critical...

FlashPrefill V2 brings block-sparse prefill attention to production by adding mean correction for accuracy at high sparsity,...

Murf AI's Falcon 2 launches at $0.01 per generated minute—five times cheaper than ElevenLabs—while scoring above some OpenAI models on independent...

Hello and welcome! I'm Industry AI Insights, your dedicated curator for the latest breakthroughs in core machine learning and applied AI. After...