Matt Wolfe walks through his Claude-based system for making weekly AI news videos. He built a prompt that turns articles and PDFs into bullet-point summaries, novelty statements, and analogies, then shows it working on a paper about an AI model that generates Doom game frames in real time.
This one is using Claude. I use cloud a lot, so a lot of my workflows use cloud. I love making projects inside of Claude, but one of these little projects that I made is called the News Summarizer. So if you watch my other YouTube channel, every Friday I put out an AI news video where I break down like, here's all of the latest AI news for the week. And this is pretty much how I come up with, like, what I'm going to say about the news for the most part is I've created this new summarizer here. And I'll go ahead and show you my system prompt. I'll read it off as well for anybody that's listening. But I have this system prompt here that says I will upload or paste in an article, PDF or video transcript. Your job is to respond with bullet points with the following information.
Summarize the content into bullet points, making it easy to understand the concepts or ideas presented. Tell me what we can do now as a result of this information that we could not do before. What makes this novel? Tell me how I should report on this in the news. If I was to give a news broadcast and explain why this information was important, what should I tell the viewers so that the most amount of people can understand it? The format of the response should be. And then I give a little format here. Summary, bullet, bullet, bullet. What makes it novel? Bullet, bullet, bullet. And then how should I describe this news or concept in a simple way that anyone can understand when I report on it? If it makes sense, use an analogy. So that's the whole system prompt there. And basically what I'm telling it to do is I'm going to give you content. It might be a PDF, might be a document, might be a copy and pasted news article. Break it down for me so I can explain it really, really easily. And one of the things that's been sort of circulating in the world of AI news this week, sure you've probably seen this, Nathan, is that Doom thing, right? The AI Doom, where it's sort of generating every single frame. So as somebody clicks the forward button, it generates the next frame. In Doom, if they click the shoot button, it automatically generates the next frame. There's no, like, underlining.
What do you call it? Like, game engine? Yeah, no, game engine underneath it. It's all being generated as they're playing. So I have that PDF and I can drag that PDF. This is literally.
The research report right off of archive.org that explains the technical concept of how this works. And if you're looking at my screen right now. If you're not looking at my screen, you can look at it over on YouTube. But I just dropped the PDF right here into Claude and I don't even add a prompt. All of my prompting was done in that system prompt that I just read. And if I just hit enter on this, it's going to read that PDF for me and then break it down with all of the information that I asked it to give me in that system prompt. So right now it's generating all that text for me here, and here's what it says. Here's a summary and analysis of the research paper. Summary. The paper introduces game engine, a neural model that can simulate the classic game Doom in real time at over 200, at over 20 frames per second on a single TPU, right? And it sort of summarizes what the article is about. And then the next section, what makes this novel interesting. This is the first neural model capable of simulating a complex 3D game like Doom in real time with high visual quality. It demonstrates the neural models can potentially replace traditional game engines for interactive 3D environments, etc. Etc. And then it goes on to say, here's how to describe this news simply, right? So now you can start to see how, like, on my AI news videos, yes, I put a lot of work into them, but, like, AI supports the hell out of me for making those news videos. So here's what it says for how to describe this news. Simply imagine if instead of programming every detail of a video game, we can just show an AI system examples of the game being played, and it can learn to recreate the entire game on its own. That's essentially what researchers have achieved with a classic game called Doom. They trained an AI to watch gameplay videos and then reproduce the game so, so accurately that even humans have trouble telling the difference between the AI version and the real thing. This could revolutionize how video games are made in the future, potentially making it much faster and easier to create complex 3D worlds and gameplay. It's like teaching an AI to be a game designer and programmer all in one, right? So that's like one of my most used workflows. And this is very relevant. We're recording this on a Thursday. I record my News videos for YouTube on a Thursday. So I was in here actually using this clod project earlier today to start to break down my news videos. Now, every piece of news that I put in those news videos doesn't need this process, right? Like, some of the news is just sort of self explanatory. But every once in a while I'll come across a research paper and I'm like, I want to go a little bit more in depth on this. I really want to make sure that the viewers understand what this means. And this is how I do that. I'll just plug in the research paper and it's going to simplify it for me and then tell me how I should break it down as a piece of news. That's awesome. I'm curious, how much time do you think this saves you? Oh, man. Well, just in term, there's a time savings, but there's also like a mental savings, Right? So it probably saves me a good 15 or 20 minutes of actually, like, going through and reading all of the concepts within the research paper. But at the same time, these research papers are often over my head. Like, let's be honest, like, once they start breaking down, like math and there's like letters within the math equation and it's like IX over 15% plus 32 AD plus 6. I'm like, you lost me with I, right? Like the very first letter in there.
AI News: Dots, GPT-6.1 Sol, Sonnet 5.5, Gemini 4, and everything you need to know
