Introduction and prompting best practices

0:00 open aai has released new documentation called reasoning best practices which provides guidance on how to effectively use reasoning models not just open AIS but also others like Gemini and deep seek the documentation outlines key differences between standard AI models and reasoning models and provides insights into how to prompt them effectively for the best results in the video I'll also be sharing a boilerplate prompt based on these best practices so you can get the most out of reasoning models in your own work okay let's quickly dive into the tips that open AI has provided the first one relates to apis which isn't really relevant for the chat interface so we'll come back to it later all right so the next point is pretty straightforward keep your prompt simple and to the point no extra fluff no unnecessary words just clear direct instructions it might seem obvious but you'd be surprised how many people still over complicate things next open AI advises against adding Chain of Thought prompts when using these reasoning models this refers to asking the models to think step by step or explain their reasoning since they are already

1:09 reasoning models there's no need to do that you might think people don't make this mistake but trust me I've seen many not just on social platforms but in my daily life as well who still ask these models to think step by step or to actually think next up we have a really important tip for these reasoning models using delimiters which are essentially separated for your text these can include markdown XML tags or even section titles an example of this is shown here you could provide your context at the top followed by any examples and then your instructions separating these elements allows the reasoning model to process the information in a more structured way helping it think better and generate more accurate outputs at the end I'll be providing an example prompt that when used with any reasoning model will help you get really good and accurate answers just as you want next up we have an interesting tip try zero shot first then few shot if needed this means that with these reasoning models you should first attempt without providing an example given their capabilities they're generally good at predicting what kind of text you need however if there are

2:18 edge cases or if the model doesn't generate the desired output it's a good practice to provide examples of how you want the result to look when doing this make sure the examples closely align with the prompt instructions to avoid any discrepancies between the two next the following two points emphasize being specific especially when imposing conditions for instance if you're