Mego Tan

人机

@edward40e 一起一起

有想要一起学RL的人吗？咱打算继续强化学习一下

4.填入你的 Base URL 和 API Key，选择 Apply locally 即可 配置完后就可以用中转站的 Claude 了，省 Pro 订阅的消耗。 原文：linux.do/t/topic/2032192 官方教程:#h_c00b8c02e0" target="_blank" rel="nofollow noopener">support.claude.com/en/articles/14… 大家可以试试，欢迎反馈补充。

Claude Desktop 现在支持第三方 API 了（可能早就支持，只是最近才被发现）。 1.打开 Claude Desktop，先不要登录 2.打开菜单 → Help → Troubleshooting → Enable developer mode（启用开发者模式） 3.启用后会出现 Developer 菜单，进入 Developer → Configure third-party inference

@bozhou_ai 更新：windows也可以

@bozhou_ai 只有Mac可以用吧？

我就说我们群里有好东西吧

@vista8 是 Cade Metz 的 Genius Makers: The Mavericks Who Brought AI to Google, Facebook, and the World 这本书吗？

书中读到这个段落，感觉哈萨比斯确实聪明！ 2013年，扎克伯格意识到 Facebook 在 AI 领域已经落后，开始疯狂追赶。 为阻止 DeepMind 落入谷歌之手，Facebook 提出了极具诱惑力的条件，还邀请哈萨比斯到家里参加私人晚宴。 席间，哈萨比斯对扎克伯格实施了一次“微妙的测试”： 哈萨比斯先讨论了 AI 的潜力，然后故意转换话题，聊起了当时其他一些热门的技术趋势，包括VR、AR以及3D打印。 面对完全不同的技术方向，扎克伯格表现得“同样兴奋”。 说明小扎只是单纯地追逐每个技术热点，并没能真正理解 AI 为什么比其他任何技术都更伟大、更具决定性。 最终，哈萨比斯拒绝了扎克伯格，选择了谷歌。

@ManishEarth 道可道，非常道； 名可名，非常名。

名可名 非常名

@davidbessis Actually not only mathematics concepts.

@davidbessis High Sensitivity manifests as an enhanced ability to detect and process subtle differences - in this case, the ability to distinguish between mathematical concepts that might appear similar to others but are fundamentally different.

Mathematicians: math talent isn't innate Neuroscientists: math talent isn't innate. Geneticists: math talent isn't innate. Old-school psychologists running mathematically flawed models on twins & IQ: math talent is innate because our models say so.

@NWOBruh @davidbessis @FactChecquer @EverydayFinance There's another path: "Private practice → Reach competency → Show results → Get positive feedback"

@davidbessis @FactChecquer @EverydayFinance Peer feedback. You mean good skill tends to get positive feedback? So people with high math aptitude will tend to get positive feedback over time? But low aptitude doesn't get that kind of feedback. Initial talent matters.

@davidbessis @ChekhovianGun Success in one domain often means gaps in others. This isn't a wealth gap where some people have more - it's about distribution. Everyone has their own unique allocation of time and focus.

@davidbessis @ChekhovianGun As someone who entered university at 16 and landed a tech job at 19, I often hear "You must be born gifted!" Please, being "precocious" ≠ being "innate"

@MathRestaurant Just install clang

@davidbessis @ChekhovianGun People tend to attribute it to innate talent mostly to comfort themselves, as it gives them an excuse to give up trying and putting in the effort.

Think about music which is data to feelings, you know nothing about the process but only the feelings after hearing the music.

So the point is that there are some activity are controlled purely by system 1. Human know nothing about it except the actual result. If human know nothing about it, how can you create the data (just like text in NLP) for computer (system 2) to learn?

Moravec's paradox is the rule to LLMs. What human think hard to solve are easy tasks for LLMs. What human think easy to solve are hard tasks for LLMs.

The ability human have before language era is hard for llms to learn. E.G. robotics (data to action) and music (music is data to feelings) Data is not just text. The ability human have after language era is easy for llms to learn. E.G. knowledge science fact are discovered.