AI News Summary 2026-08-23
.. lang: en
AI News Summary August 23, 2026
GAFAM and Major AI Companies
- OpenAI unveiled AI Futures and announced Zero Data Retention for frontier models. The first post launches a blog by the Strategic Futures team focused on preserving rights and agency as transformative AI emerges; the second describes Private Safety Processing to detect patterns of abuse without exposing content to OpenAI staff.
Influencers and Technical Blogs
- LlamaIndex published ExtractBench: The Most Comprehensive Extraction Benchmark. The benchmark is open and reproducible; it covers 370 business documents, 4,869 pages, 8 domains, 67 document types, and 14 evaluated systems, and includes Agentic Plus as a new layer of LlamaExtract.
Generative Imaging
- Microsoft AI highlighted MAI-Image-2.6 launches at No. 2 on Arena ahead of Google, Meta, and xAI. The article states that MAI-Image-2.6 ranked 2nd on Arena’s text-to-image leaderboard, 3rd in image editing, and improved by +79 Elo points compared to MAI-Image-2.5.
Chatbots and Agents
- Microsoft AI published MAI-Code-1.1-Flash: Better, faster, at a quarter of the cost. The model is described as lightweight and agentic, integrated into GitHub Copilot and VS Code, with 25% greater token efficiency and lower cost per task.
On-Premises AI and Serving
- vLLM published Exploring Speculative Decoding in vLLM on AMD GPUs. The post, dated August 23, 2026, explains speculative decoding as a draft-and-verify workflow and reports on tests conducted on AMD Instinct MI300X and MI355X using five drafting methods.