[AINews] not much happened today
Sakana AI launched a dedicated RSI Lab in Tokyo, formalizing recursive self-improvement as a research program.
“making Claude a chemist”
A daily AI digest where the standout development was Sakana AI opening an RSI Lab in Tokyo, signaling recursive self-improvement is shifting from blog-post theory to a staffed, compute-efficient research program. Secondary signals included Anthropic's Claude Mythos/Opus discussion (with some benchmark regression skepticism) and new long-horizon agent benchmarks like Agents' Last Exam mapping 1,000+ economically valuable tasks. Individually incremental, but collectively they reinforce the 'AI builds AI' narrative gaining institutional form.