In the last video, we looked at writing, reading and chatting as three major categories of tasks that you can get an LLM to do. Given that large language models were trained to repeatedly predict the next word, maybe it's no surprise that they're pretty good at writing at generating words. And it turns out that many writing tasks can be done via a web user interface. So I hope you find this video where we'll dive more into writing tasks immediately useful. For writing tasks broadly, what we typically do is start with prompts and use a relatively short prompt to write or to generate a much longer piece of text. So let's take a look at some writing applications. I often use the web interface of large language models as a brainstorming partner. If you ask it brainstorm 5 creative names for peanut butter cookies, it actually comes up with some pretty creative names Nutty Nirvana Nibbles, I would eat that. Or if you ask it to brainstorm ideas for increasing cookie sales, then it comes up with a few ideas and you can take a look to see if any of these may be useful. You can also use a large language model, again, maybe the web interface version, to write some copy for you. Let's start with an example. If you were to ask it to write a press release announcing the hire of a new COO, a new chief operating officer for your company, it may come up with a piece of text like this Company Name Welcomes, New COO's Full Name as, so on and so forth. And this is a pretty generic press release. When writing a prompt, you find that if you can give the large language model more context or more background information, then it will write more specific and better copy for you. If all that the large language model sees is this writer a press release, at this point in time, it doesn't know anything about your company, about the new CEO's name or their qualifications and so it ends up writing something very generic like this. If you end up prompting a large action model like this, it's not a problem. You may realize that you wound up with a very generic press release and decide to update the prompt to give it more information. And so if you were to prompt it and say, use the following information for the press release, this is a COO bio, this is the name of our company and some details about our company, then it will write a much more detailed and insightful press release specific to the CEOs joining this company. I find that when prompting an LLM, I'll often not get the prompt right the first time, like what we saw just now, where we had the prompt press release announcing the new hire of COO without giving any context. And that's totally fine. If you see the result isn't what you want, just revise the prompt and try again. I'll say more about this in a later video this week, when we talk about tips for writing effective prompts. Let's look at one more example. Another writing task that I sometimes use LLMs for is translation. In fact, some of the large language models you can access via Web UI are competitive and sometimes even better than the dedicated machine translation engines already, especially for languages with a lot of text on the internet. And so where the large language model has a lot of data to learn how to generate text in that particular language, it tends to do less well in languages, also called low resource languages with less text on the Internet in that language. But if you're operating a hotel and you want to translate the welcome message into formal Hindi to welcome guests, then a large language model may be able to output text like this for you. Unfortunately, I don't speak Hindi, I wish I did, but it turns out that this particular translation is only so, so the word front desk, it translates into the desk at the front rather than the reception, which is what we mean when we say the front desk of a hotel. So if you're working with a Hindi speaker, and I was when preparing the slide, then they may be able to give you some tips to say, this is some sort of not quite the best formal Hindi, but you were to tell it to translate this into formal spoken Hindi. Then it updates this text to make front desk translate into the Hindi word for reception, which is a much better translation. Now, here's one fun thing I've seen recently in the AI community, which is a lot of us that are working with translation often need to translate text into languages that we don't speak ourselves. So how can we tell if the large language model is doing something reasonable? And in fact, even if you have, say, one Hindi speaker on your team, if other members of the team don't speak Hindi, how can they figure out what's going on? So what I'm seeing multiple teams in the AI community do is translate text into pirate English for testing purposes. And so if you were to prompt a large language model to translate this into pirate English, you get, ahoy matey, we be hoping you'll relish your time aboard the Oceanview Inn. That sounds like pretty good pirate English to me. So that hobby grand worthy models be used for writing. Let's move on to peepin at Reading House.