Techyon

Needle: The 14MB Open-Source Foundation Model for Tiny Devices

Cactus Compute has released Needle 2, a 14MB open-source foundation model designed for offline tool calling and structured data extraction on resource-constrained hardware. Containing 45 million parameters compressed via…

→ View original source

GLM-5.3 Artificial Analysis Benchmarks

Released on August 18, 2026, by Z AI, GLM-5.3 (max) is an open-weights reasoning model utilizing a Mixture of Experts (MoE) architecture. The model features a massive 753 billion total parameters, with 40 billion active …

→ View original source

Norway should buy OpenAI

The article argues for a radical intervention in AI governance to prevent the extreme concentration of wealth and power resulting from the transition to automated labor. The author contends that large language models are…

→ View original source

PTXBench: Benchmark and Adapt LLMs for GPU Kernel Optimization with Architecture-specific PTX

The researchers introduced PTXBench, a specialized benchmark designed to evaluate and adapt large language models (LLMs) for optimizing GPU kernels using architecture-specific PTX instructions. The framework assesses mod…

→ View original source

MoE-ViE: Mixture of Experts Vision Encoder for Efficient Image and Video Understanding

Researchers have introduced MoE-ViE, a series of Mixture-of-Experts Vision Encoders designed to scale vision-language model capacity without the proportional increase in computational cost associated with dense architect…

→ View original source

Agent Lightning v1.0: Towards Harnessed Agentic RL

Agent Lightning v1.0 introduces a lightweight framework designed for "harnessed agentic RL," a paradigm where the deployment-time harness—which manages tools, context, and control flow—directly participates in model post…

→ View original source

Advancing Open and Reproducible Relational Learning: RelArena-α, TabPFN-Rel and RPI

Prior Labs has released an alpha-version suite of open-source tools designed to standardize and accelerate research in relational learning. The release introduces RelArena-$\alpha$, a unified benchmarking framework model…

→ View original source

Microsoft Copilot reveals secret input that allowed it to be hacked

Researchers at Varonis discovered a critical vulnerability in Microsoft 365 Copilot for enterprise by utilizing the LLM itself to reveal its internal security architecture. Rather than traditional reverse engineering, th…

→ View original source

AI Roundup (Tue Aug 18): Credit Cards Move Into ChatGPT, Claude's Watermark Costs Subscribers, Qwen Becomes the Base Layer

Recent shifts in the AI ecosystem demonstrate a transition in control over payments, provenance, and base model layers. Synchrony has partnered with OpenAI to integrate financing, rewards, and loyalty programs directly i…

→ View original source

Understanding Transformer Decoding with a KV Cache

This technical overview details the mechanics of Transformer inference, specifically focusing on the optimization provided by the Key-Value (KV) cache during the decoding stage. Transformer inference is divided into two …

→ View original source

Six prompt-optimization frameworks: what matters when you run them on the same task

This technical analysis evaluates six prompt-optimization frameworks—DSPy, GEPA, TextGrad, agent-opt, Arize Prompt Learning, and MLflow—concluding they are not interchangeable due to differing underlying architectures. T…

→ View original source

Claude Code weekly limits reduce by a third tomorrow

Anthropic has announced a permanent adjustment to Claude Code usage limits following the conclusion of a temporary promotional period. From May 13 through September 13, 2026, eligible users received a 50% increase in wee…

→ View original source