Skip to content

Sam's News — ai — 2026-08-18

AI Economics

7 AI agent workflows driving up inference costs despite cheaper per-token pricing

Despite declining inference costs, the total cost of AI agent workflows is increasing as agents perform more complex reasoning, planning, and continuous background operations.

Sources: Computerworld RSS

AI

7 EdgeRunner and U.S. Army AI Integration Center Build Army-Specific Large Language Model

EdgeRunner and the U.S. Army Artificial Intelligence Integration Center are jointly building an open-weight language model tailored for Army operations.

  • Announced August 18, 2026
  • Open-weight, locally-deployed LLM design
  • Developed for national security and Army mission-specific applications

Sources: Business Wire AI Web Searched, Yahoo Finance Web Search, Business Wire Web Search

6.5 Qwen 3.8 27B model challenges GPT-5.6 and DeepSeek V4 in benchmarks

Alibaba's Qwen 3.8 27B model demonstrates competitive performance against larger proprietary models like GPT-5.6 and DeepSeek V4 in benchmark tests.

Sources: StartupHub.ai RSS

6.5 CPU architecture experiences resurgence driven by agentic AI workloads

Agentic AI is driving renewed demand for CPU-focused architectures as enterprises seek to reduce cloud latency bottlenecks.

Sources: IEEE Spectrum Web Search

6.5 Meta Returns to Large Language Model Development After Year-Long Hiatus

Meta has resumed development of large language models after a year-long break from the field.

Sources: Understanding AI | Timothy B. Lee Web Search

6.5 DeepSeek Releases Preview of Long-Awaited V4 Large Language Model

Chinese AI startup DeepSeek released a preview of its anticipated V4 large language model.

Sources: CNBC Web Search

6.5 Large language models accelerate research software development

Large language models are disrupting the research software landscape by enabling rapid, cost-effective creation of specialized applications.

Sources: Nature Web Search

6.5 MLPerf Client v2.0 Expands AI PC Benchmarking to Image Generation and Agentic AI

The MLPerf Client v2.0 benchmark suite expands to include generative AI and agentic workflow tests for AI-capable personal computers.

Sources: Yahoo Finance Singapore RSS, Yahoo Finance Singapore Web Search

Legal AI startup Harvey releases Tenet, its first AI model designed to automate tasks traditionally performed by lawyers over hours or days.

Sources: Business Insider RSS, Business Insider Web Search

6 AI Companies Purchasing and Destroying Rare Books for Training Data

AI companies are buying rare books solely to destroy them as part of their model training and development processes.

Sources: bgr.com RSS Update to: Amazon destroys rare books to generate training data for AI large language models