AI
Artificial intelligence: LLMs, agents, diffusion, vision, NLP and the latest from top labs.
565 published articles
Edge AI
Audio8's CPU-only runtime fits voice cloning in about 1 GiB of RAM
Audio8's ONNX runtime runs the 0.6B TTS preview entirely on CPU: INT4 autoregressive weights, a bundled 44.1 kHz codec, streaming PCM, and an OpenAI-compatible endpoint. The service takes about 1 GiB of RAM on a laptop, and no PyTorch or Transformers are needed at runtime.
2026-08-05
Explainable AI
The knob no one wanted to turn: dtControl2+ε lets you sacrifice optimality for clarity
A new extension to dtControl2 allows engineers to trade a precise amount of performance for smaller, more understandable decision trees. The tool, dtControl2+ε, guarantees ε-optimality while pruning trees to orders of magnitude fewer nodes.
2026-08-05
AI image generation
Reve 2.1 hits second on Arena with a tenth of the compute
Reve 2.1 claims second place overall on the Arena leaderboards while trained on less than a tenth of the compute of the labs around it. The company also details 4K output, addressable regions, and multilingual text rendering.
2026-08-05
AI consolidation
Reve's researchers are headed to OpenAI; its products stay put
OpenAI's investment in the independent lab Reve brings a move of research talent: part of Reve's AI Research team joins OpenAI's multimodal push while Reve's products stay in place. No amount was disclosed.
2026-08-05
AI Safety: 3B Classifier, Apache 2.0 Weights
Mistral's Shieldstral puts your moderation policy in the prompt, not the weights
Shieldstral frames moderation as a binary question: an instruction, a yes/no query, and the content to judge. Mistral says the 3B model matches open guardrails up to seven times its size on text safety, with Apache 2.0 weights that run on one 16GB GPU.
2026-08-05
AI Safety
Shieldstral, the 3B classifier that outguns models seven times its size
A 3B safety classifier matches text models nearly seven times its size and sets a new multimodal moderation state of the art, per a July 2026 arXiv paper. The trick: moderation reframed as binary question answering, trained on roughly 54.1 million samples.
2026-08-05
Artificial Intelligence
LFM2.5-Encoders make the small-model case: 3.7× faster than ModernBERT on CPU
Liquid AI's open-weight LFM2.5-Encoders make the case that production NLP belongs on small models. A 230M encoder beats ModernBERT-base on benchmarks and scans full documents in about 28 seconds on a laptop CPU, roughly 3.7× faster.
2026-08-05
Compute economics
AI bought the GPUs; nobody owns keeping them busy
Procurement has a team. Utilization doesn't. With multi-gigawatt GPU deals closing across four vendors, the metric that decided airline economics is becoming AI's binding constraint, and idle clusters lose money by the calendar hour.
2026-08-04
Network Autonomy
AgentToolMO fixes a trust blind spot in multi-vendor agent networks
When Vendor B's tool is compromised, Vendor A's agents keep using it unaware. A new paper proposes a trust state machine with damped cascade propagation to bring containment from hours to near real-time.
2026-08-04
AI Research
Penelope hides its reasoning in a single decoder layer to cut inference costs
Penelope, a latent-reasoning framework for decoder-only Transformers, localizes recurrent computation to a narrow decoder interval, cutting inference latency without a major accuracy hit. The paper describes a curriculum that shifts reasoning from visible tokens to an internal GRU loop.
2026-08-04
Enterprise AI
Half of Asia's enterprises don't use English for AI. That's the bottleneck
Alibaba Cloud's NielsenIQ survey of 1,000 Asian IT decision makers finds 95% raising AI budgets and 75% calling AI indispensable. The finding that matters most: only about half of enterprises use English for AI, making local-language models the next bottleneck.
2026-08-04
AI agents
Treating SOPs as code: why compilation alone lifts strong agents by 16 points
New research from Hong Kong and mainland China demonstrates that compiling SOPs into executable pseudo-code and running them on a stack-paged virtual machine cleanly separates capable agents from brittle ones. The work yields a precise deployment rule: compile first, page only after a model-level discipline check.
2026-08-04