Logic Lab AI 🧪

A big pivot from Ken Griffin on AI: “Number one is, in the last few months, there has been a step change in the productivity of the AI toolkit. It is profoundly more powerful than it was just nine months ago. And for us at Citadel, that has allowed us to unleash a much broader array of use cases for AI. And it has been really interesting to watch, to be blunt, work that we would usually do with people with masters and PhDs in finance over the course of weeks or months being done by AI agents over the course of hours or days. These are not these are not mid-tier white collar jobs. These are like extraordinarily high skilled jobs being, I'm going to pick a word, automated by agentic AI. And I gotta tell you, I went home one Friday actually fairly depressed by this because you could just see how this was going to have such a dramatic impact on society. When you witness it in your own four walls, when you see work that used to be man years of work being done in days or weeks, it's like, wow, like that's the first time I've seen real impact in our four walls.” This echoes my own experience with agents and the conversations I am having with students, friends & clients. The toolkit has dramatically transformed and it feels like in finance, for the first time, AI is real.

About My GenAI Workflow and How People Tend to Overthink It I get asked this a lot. Bro, what's your workflow? What tools are you using? What agents? Any special prompt engineering techniques? Some specific ComfyUI node setup? And every time, I kind of feel like whatever they're imagining is this super complex, elaborate system. The reality is my workflow is dead simple. Basic. Almost embarrassingly primitive. So if you ask me what essential tools I use, my answer is just two things: 1⃣ Google Spreadsheet. Yes, literally a f**cking Google Spreadsheet. Not Notion. Not Obsidian. A freaking spreadsheet. 2⃣An LLM. In my case, Claude and Grok. That's it. That's the whole toolkit. No agents. No complicated ComfyUI nodes. No style transfer setups or whatever. Just pure prompting. And about the prompting thing, this is where I need to give some context. My prompts are 100% built with LLM. I'm no prompt wizard writing a complex prompt as if I'm a native English speaker LMAO 😂😂😂 I have a basic writing ability, but that's in my native language. Even though I understand English, it's not the language I use day to day, which means my English vocabulary is pretty limited. People who actually grew up in English speaking countries have a real advantage because they just get the nuance naturally. But me????? I don't have that. So how do I pull this off? Simple. I just use LLM. I think a lot of people still underestimate how much you can actually do a lot of things with LLM. And another thiiiing. I use zero thing of whatsoever people called it prompt engineering techniques 😆😆 What the hell is that? I see some big accounts out there sharing these super detailed instruction prompts, all structured and formatted for this specific use case or that. Well... I don't really do that 😆 The way I talk to my LLM is the same way I talk to my neighbor. "Bro, you got any ideas about this?" Or the same way I brainstorm with my creative friends. Totally conversational. I don't even set my LLM to respond in proper English. We go full rojaxxxx, mixed Bahasa Indonesia, Sunda, Jawa, Malay, Singlish, Tinglish, absolute chaos. I genuinely treat it like a friend. 😆😆 + + + Let me give you a concrete example. Here's how I approached my project Sisa Malam. I just told the LLM: "Bro, I have this idea. I want to make a story, super simple, set in a modest flat somewhere in Southeast Asia. I'm imagining the vibe being this mix of Jakarta, Bangkok, Manila..." The ideas I had were rough, coming from personal experience, things I've lived through, and my love for Southeast Asia as a region. I shared all of that, and then I asked: "What do you think?" Yes. I literally asked the LLM for its opinion. 😁 Sometimes the response was kind of lame, but occasionally interesting. The lame stuff I'd push back on immediately, same as I would with a friend "Come on bro, that's weird, why would it go like that, blabla..." And when an idea worked, I'd go" Okay yeah, that one we can use." We kept going back and forth until we hit a point where I felt: okay, this is something we can actually develop. Then we'd lock in the rules. Cinematography direction, visual approach, all of that. And again, I'd ask: "What do you think?" Because I actually want to hear what the LLM thinks. And sometimes, genuinely, the answer was brilliant. From there, I'd have a working prompt framework built in that same conversation. One where I could just describe a scene casually, like "okay in this scene this happens," and the LLM would execute it. And then comes the most important part: testing, testing, testing, and more testing. You have to test the prompt. If the result isn't working, I ask why. "Why does it look like this? Maybe we need to try it this way?" Yoooo where's the CLEAVAGE!!!! I said show the godamnn cleavage! Dude make it sexier, crash zoom-in!! LMAO!! And meanwhile!! There's this dude who's been chasing me to my email, and even to my Linkedin account asking heeey how do you bypass moderation broo??? LMAO 😂😂😂 Well i just told you how, hahahha 😆 Some even wanna pay me for consultation fee. I told them, nooooo you don't need that dude 😂😂😂 Especially with Grok, damn my conversations are pure chaos. Until eventually we hit that moment: "Okay YES. This is it. That's the one." And whatever results work well, I document everything in Google Spreadsheet. That's the process. Scene by scene. Not automated through some agent that spits out a full script. I look at each prompt one by one, we analyze together, we test together. That's the whole workflow. 100% prompting. The ideas come from me, which is what makes it feel personal. There are so many stories I was never able to express before, and now I finally can. The LLM handles the technical side of language, translating my raw ideas into something that actually works as a prompt. For the video side, I edit in CapCut. All the visual aesthetics are handled there. But how I build what goes into those visuals? That's all prompt. I do admire people who go deep into ComfyUI. The setups and workflows look genuinely impressive. But I don't go there. My approach was 100% prompt, built together with an LLM. If I want my character to do a dangdut dance, I ask the LLM how to describe dangdut movement in a prompt. No motion control. No motion reference. Just prompting. These days I'm also using Omni-Reference a lot, so most of the time I'm just building character sheets now. + + + After reading all of this, I think you can see how simple and primitive my actual process is. No complex workflows. No complex setups. 😆😆😆 And honestly, I think people overcomplicate this too much. Though I get it, it really does depend on the project. But for me, it's always this simple. And anyone can do it. The only thing you actually need is the willingness to explore. Because right now, the tendency in the AI space is that if someone sees something they like, and immediately the first question is "what's the prompt, what's the prompt?" If you take the time to actually analyze what you're seeing, you'll figure out exactly how it's done. We're all using the same tools. The difference is just how you choose to approach them. And there's no right or wrong, because what matters in the end is whether it works for you.

Customers can forgive delays. They rarely forgive dishonesty.