Prompt-Engineering-Guide VS examples

Prompt Engineering Guide is the holy grail of all guides, aiming to make it easier to stay up-to-date with prompt engineering guides, techniques, applications, and papers. If you are getting started, this is an excellent place to start.

Guides, papers, lecture, notebooks and resources for prompt engineering: https://github.com/dair-ai/Prompt-Engineering-Guide

this has been a great resource. approachable and great for practitioners. it's frequently updated with new papers and techniques https://www.promptingguide.ai/

The idea behind prompt engineering is to construct the queries given to the language models to optimise their performance. This helps to guide them to generate the desired output by fine-tuning their response. There is a plethora of research papers out there on different forms of prompt engineering. DAIR.AI published a guide on prompt engineering that you might find useful to get started.

Prompt Engineering Guide

Let's try to understand what is Prompt Engineering is all about. Here's the quote from Prompt Engineering Guide. DAIR-AI

I found this resource [0] handy for getting a grasp on all the different terms people use (zero/one-shot, tree of thoughts, RAG, etc). It's not super detailed, but was enough for me (a professional developer) to get started on some side projects with Mistral.

[0] https://www.promptingguide.ai/

There are better guides out there too

- https://www.promptingguide.ai/readings

- https://github.com/dair-ai/Prompt-Engineering-Guide/tree/mai...

- https://github.com/microsoft/promptbase (this one is less of a guide, but is likely the current SoTA)

I just published a paper detailing this non-IID check and open-sourced its code in the cleanlab package — just one line of code will check for this and many other types of issues in your dataset.

I recently published a blog introducing Datalab and an open-source Python implementation that is easy-to-use for all data types (image, text, tabular, audio, etc). For data scientists, I’ve made a quick Jupyter tutorial to run Datalab on your own data.

Cool read! I just finished up a notebook where I show how noisy labels can drastically impact the performance of Open AI LLMs. I first fine-tune the well-known Davinci model (the backbone of ChatGPT) on the original data and report an accuracy of 63%. I then use the open-source package cleanlab to find examples that are incorrectly labeled and drop them from the training data. This step increases the fine-tuning accuracy to 66% (better accuracy with less data). Finally, I correct the mislabeled examples and fine-tuning accuracy jumps to 77%!

The entire field of data-centric AI is an active field that is pretty new --- it focuses on the data side of ML as opposed to just model optimization. Our company is building an open-source package cleanlab that is becoming the DCAI standard.

I recently published a paper introducing this novel method and an open-source Python implementation that is easy-to-use for all data types (image, text, tabular, audio, etc). For data scientists, I’ve made a quick Jupyter tutorial to run ActiveLab on your own data. For ML researchers, I’ve made all of our benchmarking code available for reproducibility so you can see for yourself how effective ActiveLab is in practice.

suggest which data is most informative to (re)label next (active learning) (link)

In one-line of python, cleanlab can automatically: 1) find mislabeled data + train robust models 2) detect outliers 3) estimate consensus + annotator-quality for datasets labeled by multiple annotators 4) suggest which data is best to label or re-label next (active learning)