logoalt Hacker News

1899-12-30today at 4:45 PM0 repliesview on HN

It's interesting to note that the newer LLMs like deepseek v4 or kimi k3 basically use caveman mode natively for their thinking traces. Lot word dropping when thinking.