AI News — 2026-08-29
- Tencent unveiled Hy4 Preview, a 770-billion-parameter foundation model. A 1 million-token context window and a big jump in scale over its predecessor; Tencent claims internal-test wins over Z.AI and Moonshot. — Bloomberg
- An Anthropic fellow’s Automated Alignment Researcher improved all 10 alignment benchmarks it was pointed at. 30-minute training iterations, beating experienced human proposals within six hours at ~$4/hour of inference versus $150/hour. — TechCrunch
- A randomized four-arm trial gave 1,000+ students ChatGPT, critical-thinking training, both or neither. GPT-4o access raised the quality and coherence of their work; the causal-reasoning exercise raised idea originality instead. — OpenAI