$250K in Prizes: SafeBench Competition Announcement — LessWrong