TE
科技回声
首页24小时热榜最新最佳问答展示工作
GitHubTwitter
首页

科技回声

基于 Next.js 构建的科技新闻平台,提供全球科技新闻和讨论内容。

GitHubTwitter

首页

首页最新最佳问答展示工作

资源链接

HackerNews API原版 HackerNewsNext.js

© 2025 科技回声. 版权所有。

Hot Take: Prompt engineering isn't engineering. It's discovery

2 点作者 vapemaster超过 1 年前
I&#x27;m in the drug discovery space, and for all the &quot;engineering biology&quot; talk that&#x27;s around these days, I find a bunch of parallels to drug discovery in the AI, particularly LLM space.<p>The fact that these models get created and the poked and prodded with prompts reminds me a ton of chemical biology, aka the field of using chemicals to investigate biology.<p>It would appear that in the case of AI, the engineering has become art.<p>Ironically these technologies are the ones being heralded as the way to take biology and turn it into an engineering problem.<p>I always come back to may favorite XKCD &quot;Here to Help&quot; https:&#x2F;&#x2F;xkcd.com&#x2F;1831

2 条评论

kevindamm超过 1 年前
I think the naming originates from the similar term &quot;feature engineering&quot; which had been in use for about as long as machine learning in general. The approaches to feature engineering are also a bit of a mix of science and art -- in some cases statistical and probability theory are used to arrive at an appropriate feature mix, in other cases it&#x27;s very much throwing everything at the wall to measure stickiness.<p>Prompt engineering is definitely a bit more art than engineering, but it&#x27;s not far off conceptually from what feature engineering is.
marcus1408超过 1 年前
Also have a background in systems biology here, originally a chemical engineer. Used to do &quot;sensitivity analysis&quot; and parameter fitting on a bunch of PK models for pharma labs. Drifted to ML and now in the LLM space.<p>Prompt engineering - indeed a glorified way of plugging in values and see where comes out&#x2F;where it takes you. I remember I did not enjoy guess-timating parameter bounds on a global search, and couldn&#x27;t &quot;configure&quot; simulations as well as my experienced colleague. But well, here I am.