星期四 06 晚上 五月 16o 2024
A Front-End Engineer's Take on LLMs
A front-end engineer’s take on LLMs - For the last month or so I’ve been working on integrating the OpenAI API into a product and had my first hands-on experience with LLMs. I avoided them before because I don’t have a data science background and thought I wouldn’t have enough knowledge to use them properly. But it’s surprisingly simple and using them is more similar to writing a data pipeline than training a model. I honestly believe that prompt engineering will become a skill similar to testing. And much like we have tests living in our codebases, we’d have prompts for applications that rely on LLMs for their business logic.
Applied AI Software Engineering - RAG - Gergely Orosz’s interview with a startup explaining how RAGs work. This is one of the common patterns used to build products with LLMs and so far it’s proven to be a more cost-effective and productive way to fine tune models instead of training custom ones yourself. Definitely worth the read.
React Conf - React Conf’s first day is over and there are some great talks. The most important thing without a doubt is the new compiler. The good news are that if you’re following the rules of React so far you’re only going to get benefits from it.
Athena Crisis’ codebase - Athena Crisis is a game available on Steam and it was built entirely with React. It’s an example of great engineering and I’ve shared some of its creator’s talks in my newsletter in the past. This week he open sourced the codebase! This is great news because not only can we now run the game and make changes to it ourselves, but we can also explore the code and see how this application runs under the hood. I love reading code and I can’t wait to dig into this one.
“One of the primary reasons that abstraction is overloved is that a completed program full of the right abstractions is perfectly beautiful”
- Richard P. Gabriel
一个前端工程师在LLM的采取- 在过去的一个月里,我一直在努力将OpenAI API集成到产品中,并有我的第一个实践经验与LLMs。我以前避免了它们,因为我没有数据科学背景,并认为我不会有足够的知识来正确使用它们。但它令人惊讶地简单,使用它们更像写数据管道而不是训练模型。
应用AI软件工程 - RAG- Gergely Orosz 采访了一家初创公司,解释了 RAG 是如何工作的。这是用 LLM 来构建产品的一种常见模式,到目前为止,它已被证明是更具成本效益和生产力的方式来精简模型,而不是自己训练定制模型。
回复 Conf- React Conf的第一天结束了,有一些伟大的谈话。无疑,最重要的事情是新的编译器。好消息是,如果你遵循React的规则,到目前为止,你只会从中获益。
雅典危机的代码库- Athena Crisis是在Steam上提供的游戏,它完全是用React构建的,这是一个伟大的工程的例子,我过去在我的电子邮件中分享了它的创作者的一些谈话。本周他打开了代码库!这是一个伟大的消息,因为我们不仅现在可以运行游戏并自行进行更改,但我们还可以探索代码并看到这个应用程序如何在帽子下运行。我喜欢阅读代码,我迫不及待地挖掘这个。
“抽象被热爱的主要原因之一是,一个完整的程序,充满了正确的抽象,是完美的。
理查德·P·加布里埃尔
发布者