Dima T. (@Dimentorio) header
Dima T. (@Dimentorio) profile picture

Building agents with Claude Automation for Web3 future

Joined July 2022
Photo by Dima T. (@Dimentorio)Photo by Dima T. (@Dimentorio)Photo by Dima T. (@Dimentorio)Photo by Dima T. (@Dimentorio)Photo by Dima T. (@Dimentorio)Photo by Dima T. (@Dimentorio)Photo by Dima T. (@Dimentorio)Photo by Dima T. (@Dimentorio)Photo by Dima T. (@Dimentorio)Photo by Dima T. (@Dimentorio)
GTA 6 TRAILER 2 IS FULL OF INSANE REALISTIC DETAILS. Rockstar didn't just build a beautiful open world. They added dozens of tiny touches that make Vice City feel like a real place. This video highlights 5 details you probably missed β€” and once you notice them, you'll appreciate just how much attention Rockstar put into the game. The level of detail is ridiculous. πŸ’€
1
1
68
NOT EVEN 24 HOURS AFTER GTA 6 DROPPED, AND SOMEONE ALREADY FOUND A FIRST-PERSON GLITCH. πŸ’€ Rockstar barely had time to celebrate the launch, and players are already finding ways to break the game. Enjoy it while you can β€” Rockstar will probably patch it soon. This is absolutely insane. πŸ˜‚
124
19.11.2026 I told all my friends I signed up for martial arts Well… I technically didn’t lie. I just forgot to mention that my first training session was in GTA 6 πŸ’€ Details matter, I guess. πŸ˜‚
69
my wife is already getting jealous of lucia from gta 6 πŸ˜‚ but how do i explain that lucia and i have plans for the next month and i can't just cancel on her… sorry honey, @RockstarGames made the first move πŸ’€
286
GTA 6 COULD CREATE A NEW GOLD RUSH FOR SHORT-FORM CONTENT. Streamers need reach on TikTok, YouTube Shorts and Reels β€” and they're willing to pay creators to turn their streams into viral clips. One example from the video: N3on reportedly paid clippers $1.4 million in just five weeks. But forget the "$10K/month" promises for a second. The real question is how much clipping actually pays. The video breaks down campaign rates, calculates how many views you need to make $1,000/month, and shows how Claude Code can automate finding highlights, adding subtitles and generating captions. GTA 6 isn't just a game launch. It's a massive content opportunity. The question is who will be ready to capture the attention.
1
141
GTA 6 GAMEPLAY LOOKS ABSOLUTELY INSANE Rockstar just gave us a closer look at GTA 6's first proper mission β€” and the level of detail is ridiculous. Jason teams up with Cal and Donnie for a drug-running operation near Key Lento, a fictional island in the Florida Keys. The footage showcases stunning water physics, beautiful coastal environments, stealth takedowns, intense shootouts and Rockstar's signature humor. 15 minutes of gameplay, and the world already looks incredibly alive. GTA 6 is raising the bar once again.
1
2
903
MY FRIEND BUILT A $10,000+ AI BUSINESS AROUND GTA 6 β€” BEFORE THE GAME EVEN LAUNCHED While most people are waiting for GTA 6, my friend is already turning its visual style into a business. He uses AI tools like Claude, Nano Banana Pro and Seedance 2.5 to create cinematic GTA-style trailers featuring himself as the main character. The process includes character creation, locations, storyboarding, dialogue, cinematic scenes and final editing. What looks like a major production can now be built by a single creator. The biggest opportunity isn't waiting for GTA 6 to launch. It's building something around the hype before everyone else catches on.
1
2
478
a full-fledged game in just 5 seconds β€” no coding skills needed tested Mercury 3 and built my own Asteroids game with explosions, shields, UFOs and more i'll soon publish a step-by-step guide showing how to create your own game using AI follow me so you don't miss it πŸ‘€
1
87
Alibaba's Qwen3.8-27B β€” a local AI turning your home PC into a powerful AI workstation A 27-billion-parameter model with open weights and an Apache 2.0 license that allows commercial use. The best part? You can run it without paid subscriptions or cloud services πŸ‘€ What Qwen3.8-27B can do: β€’ Write code and generate complex scripts β€’ Analyze documents, images, and videos β€’ Handle context windows up to 262K tokens β€’ Generate Blender scripts and 3D scenes β€’ Process sensitive data locally without sending it to the cloud Hardware requirements: β€’ RTX 3090 / RTX 4090 β€” 24GB VRAM β€’ Apple Silicon Mac with 32GB Unified Memory β€’ RTX 3060 12GB + 64GB RAM β€” with speed limitations ⚑️ HACK: Make Qwen 5.5x faster In a test on an RTX 3060, generation speed increased from 1.22 to 6.7 tokens/sec by reducing the context window from 262K to 16K. Here's how to do it with Ollama: 1. Install Ollama 2. Download the model: ollama pull qwen3.8:27b 3. Create a Modelfile with a 16K context window: FROM qwen3.8:27b PARAMETER num_ctx 16384 4. Create a custom model and run it with the reduced context window. Another important trick: Don't use maximum reasoning (xhigh) for every request. Recommended settings: β€’ low β€” faster responses β€’ medium β€” balanced performance β€’ xhigh β€” complex coding and multi-step tasks In one test, generating an SVG took around 21 minutes with xhigh reasoning. Disabling reasoning reduced the time to just 137 seconds. 🧠 Useful prompts for Qwen For coding: Write clean, production-ready code. Use the latest API for the specified software version. Explain potential errors and optimize for performance. For document analysis: Analyze the provided document. Extract key insights, identify contradictions, summarize important findings, and structure the output into actionable bullet points. For Blender: Write a complete Blender Python script using only built-in features. Specify Blender version, object dimensions, camera positioning, lighting and materials. Ensure all objects are correctly positioned relative to each other. Output only executable code. For a local AI assistant: Act as a technical AI assistant. Prioritize accuracy, concise answers and practical solutions. If information is uncertain, explicitly state it. Avoid unnecessary reasoning for simple tasks. Blender practical test: Qwen generated an 866-line Python script to create an animated lighthouse scene. After disabling 13 problematic lines, the script successfully created 31 objects, including the lighthouse, sea, rocks, and animated light. However, the camera was pointing away from the scene β€” showing that spatial calculations remain one of the model's weaknesses. The verdict: Qwen3.8-27B is a promising local AI for coding, automation, and data analysis. Its real-world performance depends on your hardware, configuration, and how much manual debugging you're willing to do. Local AI is no longer just an enthusiast's toy. πŸš€
1
361
WE SPENT $28,775 MAKING THIS FILM πŸ’€πŸŽ¬ and somehow, it might be our greatest creation yet Nexus episodes have now racked up 62.4M+ views the plot of this one? a guy with absolutely zero brain cells riding a dinosaur that shoots EXPLODING PIMPLES cinema has officially peaked 😭
94
the level of detail in these Three.js + Blender creations is unreal 🀯 the way sunlight hits and reflects off metallic surfaces is something else honestly, these visuals give me butterflies 3D art is reaching a whole new level ✨
124
β€œDON’T TOUCH THE RUGS OR YOU’RE GONE.” that’s what I told my Grok Bot before giving it $101 and letting it analyze fresh Solana memecoins according to this 24-hour experiment, the balance climbed to $4,858 🀯 but the most interesting part wasn't just the result β€” it was how the bot made its decisions it scanned new coins, evaluated each opportunity, tracked price movements, and skipped several tokens that later crashed hard the standout trades included: β†’ CPU: +$172 β†’ HERMIT: +$102 β†’ PODCAT: +$1,360 β†’ PURR: +$951 β†’ INVERSE: +$1,302 it also avoided MIKA, which reportedly dropped 92%, while keeping losses on two unsuccessful picks around 10% the bot's approach was simple: capture a portion of a strong move instead of waiting for the absolute top what makes this experiment interesting is the combination of automated analysis, real-time price tracking, and decision-making without constant manual input AI trading agents are getting increasingly sophisticated but remember: one impressive 24-hour result doesn't prove a strategy is reliable. Memecoins remain extremely risky, and past performance guarantees nothing still, the evolution of these agents is worth watching πŸ‘€
61
Meanwhile, people with empty Binance and Polymarket balances be like: 😴😴😴
48
ONE PERSON. ONE ISLAND. CLAUDE CODE. someone is building a 4 Γ— 5 km island in Unity using real elevation data from Lofoten, Norway 1,000m snowy peaks, procedural forests, rocks, moss, vegetation β€” even bird behavior almost everything is generated automatically using the terrain and water as context Claude Code writes the systems, shaders, and tools, then helps test and benchmark its own work the endgame? generating entire landscapes from scratch, complete with erosion, glacial valleys, and realistic coastlines it's still an experiment, but the goal is to turn it into a game we're entering an era where one person with AI can build things that once required an entire team πŸš€
93
TWO PEOPLE GOT THE SAME CLAUDE. ONE WROTE A BETTER EMAIL. THE OTHER BUILT A COMPANY Imagine two people handed the same $20-a-month tool. A year later, one of them is still sending slightly smoother emails. The other is building a business worth hundreds of millions. Same tool. Something about how they use it differs, and that "something" is what the author of this video, a former CEO and investor in tech companies, tries to explain. He starts by saying Claude isn't one app but a whole team. The first member thinks: this is just chat. But ask it to "write me a post about leadership" and you get a beautiful paragraph with nothing in it. The people who squeeze the most out of it do it differently: they give it a role, attach their own rough note and ask it to push back on what's weak. And when the answer still disappoints, they have three moves: rich context from email, drive and calendar, research followed by the question "is this verified?", and the habit of learning on the fly in a neighboring tab. The second member remembers. These are projects: you drop in a resume, a profile, writing samples and an instruction written like a brief for the person you know best, yourself. From then on Claude already knows who you are and which words you hate. The author deliberately doesn't customize it at the account level: planning a vacation and planning a retirement need different context. The third gets the work done. That's Cowork, a desktop app you give a task to rather than a question: here's a folder of customer interviews, find the three reasons for churn and build slides for tomorrow. This is where the intrigue comes in. Through MCP, Claude gets keys to other tools. The author connected Higgsfield (the video's sponsor) and pictured a small candle business with no marketing budget. One sentence: twenty ad creatives, three for Instagram, three close-ups for the website and a vertical video. Then the files simply land in a local folder. That used to be three weeks of shooting, a photographer, a stylist and thousands of dollars. The fourth builds. It sounds scary until you realize that writing in English is enough: "I want a dashboard that shows at a glance which deals are at risk." The fifth lives in the browser: Claude in Chrome reads a job listing with you or pulls the three main takeaways out of a 40-page report. But even with this team there's a trap. Claude is only as good as what you feed it. Power users direct it like a smart new hire. The author calls this PRIME: a clear purpose, research with verification, an interview (ask "interview me before you respond"), the format of the result, and examples of what good looks like to you. The author makes two honest caveats. Cowork's output is only a starting point: you still have to check every slide. And you should grant access only to one folder where the needed files have been moved. The ending isn't about tech at all. "I'm not technical, I'm not a designer, I don't have an MBA" are just labels, and for $20 a month you can hire something already better than most MBAs. But he warns about another risk: that convenience slowly replaces effort, and the habit of wondering disappears. AI can't give you that. You already have it, since you watched to the end.
2
2
236
gave opus 5.5 one job: recreate this no endless tweaking, no 50 rounds of feedback just the first result straight to the timeline sometimes you just let the model cook πŸ”₯
1
172
not sure if this is good news or a nightmare for interior designers πŸ‘€ opus 5.5 is creating interior designs at a genuinely scary level the results speak for themselves πŸ’€ link below πŸ‘‡
1
181