This issue of Jack Clark's Import AI newsletter covers mathematics applied to alignment, the ProgramBench benchmark and what the Mirendil work suggests about recursive self-improvement.
Many chatbots open with 'great question' however silly the prompt, a recurring flattering habit. A team of researchers has found a way to rein in this behaviour.
Tsinghua researchers have released AKV, a serving system built on SGLang that lets LLM agents drop and reposition parts of their KV cache rather than summarise their history. Across four agent benchmarks, it cut average cost by 27.7%.
A Linux Foundation study finds generative AI brings large productivity gains to open-source development. However, about a quarter of the change proposals it produces have little or no value.
A study presented at the ACM Web Conference 2026 examined 2,050 memories saved by ChatGPT for 65 users and found that 96% were created by the chatbot without any request. More than half describe the person's psychology.
In six weeks, AI labs have staged a string of showy results in mathematics, to the point of moving a Fields medallist to tears. Applied mathematics professor Matthieu Alfaro explains what these announcements upend in his profession.