对 LLM 的虚假宣传太夸张了

来自这个报告的图片

GPT4 做简单算法题,宣传:全对。

实际测 div2-div4 的 problem A,18 年的题目全对,23 年没见过的【全错】。

测试集 \subset 训练集,怪不得宣传这么牛逼呢

2 Likes

AI is very stupid. It only remembers everything you feed it. You have seen the solutions before you take the exam, you will probably be the No. 1. That’s why I am pessimistic about AI, machine learning, LLM and everything in this scope. Conducting such research does NOT make ANY sense. I have seen many artificial tools that regurgitate information, but I haven’t seen even one that has logical mind. I don’t like people who are doing spurious AI research just apply the model developed by others to the fields they previously worked on. That’s industry, not academia. Do you believe it? They even get these suck papers published in top conferences. How ridiculous.

3 Likes

逻辑推理确实 shit,但在某些任务确实是最佳选择,比如搜索引擎,机器翻译初稿,语义检索,破防柯洁

3 Likes

赚的是投资人的钱。
没有东西可以投资把钱扔在账面上可能更让投资人恐慌

3 Likes

毕竟 2020s 特色焦虑经济,谁掌握了焦虑谁就掌握了财富。

2 Likes

哈哈,比如月之暗面还没盈利就开始套现了。凭空变出来一堆钱。

4 Likes

https://cn.wsj.com/articles/尚无业务的ai初创公司为何能拥有大量现金-3d9e4308