Ask HN: Is post-training make AI models dumber?
Post-training is making models better at benchmarks. They seem more consistent at coding, but dumber to me and I want to know what y'all are noticing. Here are the big three I'm noticing (mostly in claude): 1) Agents cannot write natural language well. Claude uses vague phrases to avoid any technical term such as "What the caller calls" instead of saying "callee". 2) Agents cannot manage encapsulation; they always make spaghetti code. 3) Agents are afraid to break anything. T
阅读原始来源
来源记录
指标快照
points · 2 votes
2026-11-15 04:11:10.841Z
cumulative since publication
研究分析
用户报告的问题
来源描述：Post-training is making models better at benchmarks. They seem more consistent at coding, but dumber to me and I want to know what y'all are noticing. Here are the big three I'm noticing (mostly in claude): 1) Agents cannot write natural language well. Claude uses vague phrases to avoid any technical term such as "What the caller calls" instead of saying "callee". 2) Agents cannot manage encapsulation; they always make spaghetti code. 3) Agents are afraid to break anything. T
频次与替代方案
记录作者：CleverLikeAnOx。 来源票数: 2 votes (2026年11月15日 UTC; cumulative since publication) 可供比较的近期同类记录：“Ask HN: What AI agent infrastructure are you using in 2026?” (Hacker News, 2026年11月15日)；“[FeatReq] Please create a SKILL for letting agents using transformers” (GitHub Issues, 2026年11月14日)。主题相近仅用于导航，未核验其是否属于同一需求或独立证据。
需求判断
将报告的任务或障碍作为访谈线索。功能请求、抱怨或替代方案搜索本身并不能确定预算负责人或付费意愿。
商业证据缺口
待补证据：受影响的独立用户、重复发生场景、当前绕行方案、实际时间或金钱成本、采购权限，以及对具体报价的响应。
需求验证动作
按相同任务整理原始问题，去重作者和转载；核对今天如何处理、实际成本多少，再测试明确报价后判断付费需求。