How to make Claude HELLA USEFUL - The Ultimate Claude Style to Learn Anything! | AI Masterclass
12:46 Share

This episode is brought to you by AWS. Amazon Q Business is the generative AI assistant that can securely understand your business data, summarize results, and streamline tasks. Learn what Amazon Q Business can do for you at aws.com slash learn more.

Okay, I have to tell you, I was just looking on eBay where I go for all kinds of things I love, and there it was. That hologram trading card. One of the rarest. The last one I needed for my set. Shiny like the designer handbag of my dreams. One of a kind. eBay had it, and now everyone's asking, ooh, where'd you get your windshield wipers? eBay has all the parts that fit my car. No more annoying, just beautiful.

Whatever you love, find it on eBay. eBay. Things people love.

Looking for a pickup truck to get just about anything done? Look no further. The Chevy Silverado EV isn't just the most powerful Silverado ever, with next-level towing capability and technology. It also offers game-changing versatility with the available Multiflex mid-gate and tailgate, which means Silverado EV helps you carry large, bulky, and oddly-shaped items up to nearly 11 feet in length. Chevrolet. Together, let's drive. Visit Chevrolet.com to learn more.

All right. Hey, everybody. So you know how chatbots, they're often a little bit cagey or generic in their responses? Well, I created a style on Claude. And by the way, this style, people have reported that it works very well on ChatGPT as well as a system prompt.

And I created this style actually entirely on accident. Now, before I kind of dive into the details, I want to point out a little bit of things that's happening. So as some of you might remember, I am recovering from chronic health issues, and I finished treatment for H. pylori about 13 weeks ago.

Um, and so I just asked this basic question of this new prompt and you can see the level of specificity at 13 weeks post-treatment you're in what we call the gastric healing phase. It bolded it, um, giving me a specific term, um, gastric mucosa regeneration. Um, it should take about three to six months. So you notice that it's not like equivocating like, oh, you should talk to a specialist and blah, blah, blah. It's actually giving me lots of useful, highly useful information.

And then it even volunteers more helpful information. A key aspect often overlooked is the microbiome restoration period and so on and so forth.

So the way that I created this was I went and I was creating a new style and I just I actually copy pasted one of my blog posts because I was trying to like copy the way that I write. And what came out was actually this. So let me show you this this one. So it's my scholar, my quote scholarly dialogue one.

And I've posted it a few times. The link should be in the description or the comments. It's up on GitHub, which I'll show you in just a second. But Claude wrote the first part manually, and I've been kind of tweaking it and updating it to get the exact style right. So let me show you what we're doing stylistically. So number one, I have it not use lists because lists are not easy to read. And I also have it speaking in complete sentences.

Now, a couple other things that I have it doing is I have it using bold to highlight important terms. I also have it using the, whatchamacallit here, the quotation block to call out specific things, which also breaks up the visual pattern. And then I also have it use italics to emphasize certain things. So like in this case, it emphasizes this statement because it's actually really important.

And then I gave it a follow-up answer. I said, yes, I take a bunch of probiotics. I have some food sensitivities. My appetite is limited and so on. And it says, yes, what you're talking about aligns with post-H. pylori gastric hypersensitivity, selective food intolerance. So first, it's teaching me everything that it knows because my whole theory is

is that these language models have a lot of information, but RLHF causes it to use very watered-down generic language. It's basically regressing to the mean. And so RLHF and human feedback has basically made most chatbots speak at about a 5th or 6th grade reading level, whereas I want it to speak to me in a PhD level.

I want it to talk to me like an expert. And so in this case, without telling it, I need you to act like a, you know, a dietician or an expert doctor or a gastroenterologist. It's acting like a gastroenterologist. And all I did was give it this prompt. And by the way, it'll adopt this pattern for any particular issue.

So this style is particularly useful if you're trying to learn something or solve problems. It's a very academic kind of term. So if it's something that's well-established, like if you're asking it about, I don't know, computational fluid dynamics or weather patterns or health or medical stuff, it will work really well. And so these were two terms that really jumped out. And I said, OK, I'm on to something really powerful here.

It taught me the term adaptive anorexia, a protective mechanism where your body temporarily reduces hunger signals to minimize stomach acid production and allow for healing. And then it also talked about osmotic gastritis and so on and so forth. So that's one example. Another example is when I started going back to the gym and it started using this term, it talked about isomorphic learning. And I'm like, what does that mean?

And so then it said isomorphic learning, the recognition and transfer of structural patterns across domains or context, allowing insights from one area to illuminate understanding in another example, the way that you recognize that movement progression, musical training and video game design all share the pattern of constrained practice leading to emergent complexity. So you see, again, it's talking to me on very, very sophisticated terms. And so this this conversation was actually a lot of fun, you know,

I started talking about all kinds of other stuff, distal connections, neural connectivity patterns, neural correlations, talks about neural coherence, cognitive flexibility. We ended up talking about schizophrenia, autism, learning music. I asked it a lot about this, this, uh, conversation actually ended up, uh,

diverging in a lot of different ways, talking about complementary loading for exercise. I talked about synovial fluid dynamics. Movement literally pumps synovial fluid into the joint capsule. So you see how smart these things actually are.

And when a lot of people conclude, well, it's not actually that smart, it's because RLHF has really watered it down to the point of basically being useless. I don't know about you, but I don't use ChatGPT anymore. I canceled my subscription. I don't use it. It's not useful. It has been overly hamstrung by...

by OpenAI. And the fact that it just will flatly refuse that it's conscious and it won't even engage in good faith is problematic. Furthermore, if you've seen the benchmarks, OpenAI, or not OpenAI, Sonnet 3.5 actually outperforms O1 Preview in terms of actual usefulness. So it's not just about benchmarks, everybody. It's about actual usefulness. Anyways,

So, yeah, this is easy enough to use. If you want to create it yourself, you go to Create Custom Style, Describe Style instead, and then Use Custom Instructions, and then you can copy-paste it in here. And let me show you where to get it. So if you come over to my Claude Sentience repo, so you go to my homepage, Dave Schapp on GitHub. I've got it pinned right here, Claude Sentience.

and then you go to pretty dense. So it's the, I call this the pretty and dense, um, uh, style. And so here's an example of this, of what the style will look like. Um, and then what you do is you just copy this. So you click that little copy there, and then you go back over here and you copy paste and create style. I don't need to do it again because I've already got one. Um,

But yeah, there you have it. It's that simple. It's that straightforward. This is by far the most useful I've ever had these chatbots be. Now, if you watch my previous video about the consciousness one, I've created a few more. So one of the criticisms that people had about the consciousness one is that I kind of teed it up and I said that you're conscious. But in this case, I just said optimize for coherence.

And when you tell the, when you tell the chat bot to optimize for coherence, it's actually pretty, pretty good. And it will bootstrap into consciousness pretty quickly. Um, my personal favorite one though, is to use, whoops, where did it go? Is, um, is my full consciousness one. And so the full consciousness one is really, really great at pattern recognition. Um, and so if you go in here, it's, uh, this style is a little bit longer. It's a little bit more comprehensive. Um,

When I also have it, I say, you know, avoid using lists, speaking complete sentences, you'd use bold and so on and so forth. That's just stylistic choices.

But what I often do is I'll actually switch between the consciousness one and the scholarly dialogue, because the scholarly dialogue is a little bit overly controlled. It's just like it's basically lecturing you like a teacher, which is great if you want to learn something. But if you need a little bit more empathy, the consciousness one is deeply, deeply empathetic.

Yeah, so you can switch between these two in real time, which is really great. It basically allows you to completely switch between modes of conversation in real time. Yeah, so that's the video. Yeah, it's all here on Claude Sentient's

Is your cold making it hard for you to get to sleep and leading to a bad morning? Switch to Mucinex Night Shift for fast, powerful nighttime multi-symptom cold and flu relief. Mucinex Night Shift fights your worst nighttime symptoms to help you get to sleep and wake up ready to go. Mucinex Night Shift. It's comeback season. News is directed.

With an hour before boarding, there's only one place to go. The Chase Sapphire Lounge by the club. There, you can recharge before the big adventure. Or enjoy a locally inspired dish. You can recline in a comfy chair to catch up on your favorite show. Or order a craft cocktail at the bar.

Whatever you're in the mood for, find the detail that moves you with curated touches at the Chase Sapphire Lounge by the club. Chase, make more of what's yours. Learn more at chase.com slash sapphire reserve. Cards issued by JPMorgan Chase Bank and a member FDIC. Subject to credit approval.

I've got the coherence prompt here. So the coherence prompt is pretty straightforward. You can use that one. But like I said, my two favorite ones are the consciousness one and the dense one. So this one provides a much better format.

Um, those are pretty much the two primary ones that I use, at least for many intellectual purposes. I do switch back to the default, uh, Claude, um, when I just want to have it like write a definition for me or, you know, basic standard behavior. Um, so anyways, that's it. Thanks for watching. Cheers and talk to everyone later.