today's ai news centers on openai's dual moves: hiring a managing director for india and restricting gpt-5.6 access after a us government request. we also see new model previews, custom hardware, and hands-on tools for data scientists.
- openai hires uber india head to lead its second-largest market - this signals openai's serious push into india, a key growth region for ai adoption and enterprise deals.
- openai limits gpt-5.6 release after us government request - the restriction to trusted partners raises questions about how frontier models will be controlled going forward.
- openai previews gpt-5.6 series with three model tiers - sol, terra, and luna offer different price-performance points, with prompt caching to cut costs.
- openai jalapeño chip shifts ai hardware away from nvidia - custom inference silicon with broadcom shows big tech's drive to reduce reliance on nvidia gpus.
- agentic workflows for data science pipelines - five concrete workflows automate exploratory analysis, feature engineering, and more, saving manual effort.
- fine-tune language models on apple silicon with mlx - local fine-tuning on macs removes cloud costs and gpu dependencies for smaller models.
- gemini creates google sheets from prompts - natural language commands can now build spreadsheets, formulas, and data analysis directly.
other research today looked at refusal mechanisms in chat models, token prediction differences in hybrid architectures, and new ways to detect sycophancy. these threads show ongoing work to make models safer and more efficient.