WEBVTT 00:00.860 --> 00:03.530 Okay, let's walk through anthropic Claude. 00:03.530 --> 00:06.710 It's, uh, my current favorite model, actually. 00:06.710 --> 00:08.540 So it's a transformer model. 00:08.540 --> 00:09.620 Just ChatGPT. 00:09.830 --> 00:17.840 And actually, the team that created Anthropic's Claude model, I think, originally came from OpenAI 00:17.870 --> 00:18.320 as well. 00:18.350 --> 00:20.030 Yeah, they left because of safety concerns. 00:20.030 --> 00:24.590 And you could definitely find that you get a lot more refusals with this model for certain things. 00:24.680 --> 00:28.010 That's one caveat, but that said, it feels like the smartest model. 00:28.010 --> 00:30.860 So currently I'm using sonnet 3.5. 00:30.890 --> 00:32.540 It might be a newer model out there. 00:32.540 --> 00:38.120 Maybe they haven't retained their edge by the time you're watching this, but right now it's a really 00:38.120 --> 00:44.540 great model and it's accessible via the cloud console or the API, as well as through a web app and 00:44.540 --> 00:45.920 a mobile app as well. 00:46.160 --> 00:52.370 What major difference in terms of approach is that anthropic seems to focus more on constitutional AI 00:52.370 --> 00:59.660 or synthetic data, like training it based on a list of rules, in order to get it to follow those rules 00:59.720 --> 01:07.110 rather than using Rlf, which is what OpenAI used for its training it based on individual preferences 01:07.110 --> 01:07.830 from people. 01:07.830 --> 01:13.440 So in terms of performance, I would say that this is artificially low. 01:13.470 --> 01:14.940 It's definitely not seventh. 01:14.970 --> 01:18.300 It's definitely it's definitely way better than GPT four or mini. 01:18.330 --> 01:24.000 In my experience, I think the reason why the the rankings are a little bit lower for it is because 01:24.000 --> 01:25.470 it has so many refusals. 01:25.500 --> 01:28.200 I will just say no to some types of tasks. 01:28.230 --> 01:32.340 And also maybe the formatting isn't as perfect as like what people would want. 01:32.370 --> 01:37.140 One of the things I find is that it gives you a bit of a preamble of the text, and a little bit of 01:37.170 --> 01:39.090 a description of what it's done. 01:39.120 --> 01:40.530 It can't really help itself. 01:40.740 --> 01:45.000 So you have to work a little bit harder to parse the answer in my experience. 01:45.030 --> 01:52.230 Now, in terms of features that make it really powerful, one of the first things I did was I uploaded 01:52.230 --> 01:56.040 my whole book into the prompts, and it actually takes it. 01:56.070 --> 02:01.450 They have a at the time that 100 K contact center, now it's 200 K might even be longer. 02:01.480 --> 02:07.090 So I think only Google has a longer context window, but it does a really good job of document understanding 02:07.090 --> 02:08.140 because of that. 02:08.380 --> 02:10.360 It's also a great creative writing partner. 02:10.390 --> 02:15.010 Like, you can throw in a bunch of examples of something you're trying to do, like here I've got a 02:15.010 --> 02:21.880 bunch of examples of my blog post titles, and then it's coming up with good other ideas as well. 02:21.910 --> 02:26.410 It feels more creative than the other models, like some of these names are great, like the prompt, 02:26.410 --> 02:29.140 Whisperer, prompts and Wonderland. 02:29.170 --> 02:30.880 Dancing with Digital Socrates. 02:30.910 --> 02:32.290 Maybe a bit too crazy. 02:32.320 --> 02:33.970 And then Code Generation. 02:33.970 --> 02:35.590 It's currently the best model. 02:35.620 --> 02:40.630 The one that most people use for writing code in cursor specifically. 02:40.660 --> 02:45.910 You can choose which model you use when you're using the cursor code editor. 02:46.180 --> 02:48.880 This is like their version of GitHub Copilot. 02:48.880 --> 02:51.340 And man does it work really well. 02:51.370 --> 02:52.960 It's really useful. 02:53.080 --> 02:57.790 That's tends to be what people use it for document understanding, creative writing, and code generation. 02:57.790 --> 02:58.870 So try this model out. 02:58.870 --> 03:00.580 And yeah, let me know if you like it.