The week ahead: 20260302

Industry

• [Blog] Amp self-destructs.
• [News] Goldman Sachs to launch SPXXAI: the S&P 500 index without anything AI.
• [Tweet] Polymarket API for agents.

Paper

• [arXiv] LLM Generated Persona is a Promise with a Catch: we can’t synthesize human personas with LLMs in the superficial way.
• [arXiv] AlphaEvolve to discover unintuitive new algorithms.
• [arXiv] Reasoning models are only better than standard models on problems of medium complexity. On simpler problems, they overthink; on harder problems, they collapse.
• [Blog] RL basics.

Other

• [GitHub] BullshitBench: measure AI’s ability to push back on nonsense.
• [Tweet] The harness is the product. The scaffold matters more than the model.
• [GitHub] Automaton: an agentic loop that manages its own budget.
• [Tweet] Apps are disappearing.
• [博客] 模型加上四個工具就可等同OpenClaw。

Have a great week!