WEBVTT 00:00.330 --> 00:01.800 -: Okay, so this is just a quick video 00:01.800 --> 00:03.300 showing you the different types of things 00:03.300 --> 00:04.530 that you might want to look at 00:04.530 --> 00:06.780 when you're exploring OpenAI's platform. 00:06.780 --> 00:08.910 If you click on the Dashboard in the top right, 00:08.910 --> 00:11.310 you'll see that you have access to logs. 00:11.310 --> 00:12.720 This can be a way for you to see 00:12.720 --> 00:15.180 how your application is performing in real time. 00:15.180 --> 00:18.000 This is something that when you have a productionized app, 00:18.000 --> 00:19.980 you will look at the logs and see exactly 00:19.980 --> 00:22.230 how the responses are coming out versus users 00:22.230 --> 00:25.500 that are using your chatbot or your LLM-based solution. 00:25.500 --> 00:28.620 There's also traces, which is specifically for agents, 00:28.620 --> 00:30.570 and you also have batches. 00:30.570 --> 00:32.010 I'm gonna skip over assistants 00:32.010 --> 00:33.930 because that will be deprecated. 00:33.930 --> 00:38.520 Batches allow you to do a 50% cost reduction against OpenAI 00:38.520 --> 00:40.650 with a 24-hour turnaround. 00:40.650 --> 00:43.290 This is what it would look like in a real life project 00:43.290 --> 00:44.730 where you have these batch files, 00:44.730 --> 00:46.830 you can go and look at the input and the output, 00:46.830 --> 00:48.900 and you can see whether the batch was canceled 00:48.900 --> 00:51.060 and what the status of that batch was. 00:51.060 --> 00:54.060 The evaluation section allows you to create evaluations 00:54.060 --> 00:56.250 for different types of your models. 00:56.250 --> 00:58.080 So the way that this works is 00:58.080 --> 01:00.060 you store your chat completions, 01:00.060 --> 01:03.510 you create data for the input and the prompts manually, 01:03.510 --> 01:07.680 you upload a file, and then you create a custom evaluation 01:07.680 --> 01:10.260 where you basically will be able to figure out 01:10.260 --> 01:11.880 whether your chatbot is actually doing 01:11.880 --> 01:13.620 what it thinks it should be doing. 01:13.620 --> 01:16.350 Fine-tuning is where you can create custom models 01:16.350 --> 01:18.210 based on your training data. 01:18.210 --> 01:19.740 You can customize these models so that 01:19.740 --> 01:22.200 you can easily have more control. 01:22.200 --> 01:23.910 I would recommend looking at this 01:23.910 --> 01:27.480 when you have around 1,000 to 2,000 examples 01:27.480 --> 01:30.210 and your application has been running for several weeks 01:30.210 --> 01:32.160 and you have lots of data. 01:32.160 --> 01:36.060 You also have access to storage, so files and vector stores. 01:36.060 --> 01:40.050 Vector stores allow you to basically store some text, 01:40.050 --> 01:42.420 which you can then search against in real time, 01:42.420 --> 01:45.210 and you will often use files which are stored 01:45.210 --> 01:47.850 inside of vector stores, but you also use files 01:47.850 --> 01:49.380 for the batch API too. 01:49.380 --> 01:52.230 The other good page to look at is something called usage, 01:52.230 --> 01:53.820 and the usage, if we're gonna go 01:53.820 --> 01:55.890 and have a look at one that has more, 01:55.890 --> 01:57.390 we'll show you the breakdown. 01:57.390 --> 01:58.320 So you can see here we have 01:58.320 --> 02:00.060 a large number of chat completions. 02:00.060 --> 02:01.830 If we look at the spend categories, 02:01.830 --> 02:04.140 you can see which type of models you have 02:04.140 --> 02:06.090 and what costs of those they are. 02:06.090 --> 02:08.160 If we go back to the API capabilities, 02:08.160 --> 02:09.480 you can see we're mainly spending on 02:09.480 --> 02:10.980 the chat completions endpoint. 02:10.980 --> 02:13.530 However, there's a variety of different services 02:13.530 --> 02:15.960 that OpenAI offers, and you can see 02:15.960 --> 02:18.480 how much you would spend on that. 02:18.480 --> 02:22.110 You can also filter this by the 14 days or the 30 days, 02:22.110 --> 02:25.200 and you can see the breakdown of users and services 02:25.200 --> 02:26.880 and API keys as well. 02:26.880 --> 02:29.040 The other things that I'll just point out here is 02:29.040 --> 02:31.560 there's this Cookbook link that is a really good place 02:31.560 --> 02:33.480 for you to figure out how to do different types 02:33.480 --> 02:35.730 of applications using OpenAI. 02:35.730 --> 02:37.590 As well as that, I would also recommend 02:37.590 --> 02:39.480 in the top right clicking on Docs 02:39.480 --> 02:42.120 and having a read through all of the core concepts, 02:42.120 --> 02:44.970 the built-in tools, and also agents. 02:44.970 --> 02:46.590 This is gonna give you a good understanding 02:46.590 --> 02:48.870 as to all the different types of capabilities 02:48.870 --> 02:51.333 that OpenAI's API offers for you.