Self-improving LMs are becoming reality with MIT's updated SEAL technique | Dark Hacker News