vanvster (@vanvster) header
vanvster (@vanvster) profile picture

Ω head chef at the AI

Joined April 2023
Photo by vanvster (@vanvster)Photo by vanvster (@vanvster)Photo by vanvster (@vanvster)Photo by vanvster (@vanvster)Photo by vanvster (@vanvster)Photo by vanvster (@vanvster)Photo by vanvster (@vanvster)Photo by vanvster (@vanvster)Photo by vanvster (@vanvster)Photo by vanvster (@vanvster)
vanvster retweeted
Seven fixes that stop your posts from sounding like ChatGPT. Steal the list: 1. Cut the hype words: seamless, effortless, game-changing, revolutionary. Replace each with something a reader could check - "sets up in 10 minutes" is better than "effortless". 2. Cut the stock phrases: "in today's fast-paced world", "say goodbye to", "take it to the next level". Say the specific thing instead. 3. Kill the pivot: "it's not X, it's Y". Say what the thing is, plainly. 4. Watch the long dashes (—). AI overuses them. Split the sentence or use a comma instead. 5. Delete the chatbot residue: "Let's dive in", "In conclusion", "I hope this helps", and any opener that announces the post instead of starting it. 6. Keep one hedge at most. "Could potentially" and "may eventually" cancel each other. If you're genuinely unsure, say so once. 7. Vary the shape. Equal-length paragraphs and same-weight sentences read AI-written even when every word is clean. And one thing not to fix: your quirks. Odd word choices and uneven rhythm are the proof a person wrote it. When in doubt, leave your sentence alone. All seven checks run in one free command: /gtm humanize in Adaptico OS - open source, runs in Claude Code. 📌Get the free skill here: github.com/adaptico/adaptico…
5
6
7
26
14,325
vanvster retweeted
THE MOST POWERFUL AI EVER BUILT IS LIVE RIGHT NOW. YOU'RE NOT ALLOWED TO USE IT Anthropic shipped Mythos. OpenAI shipped GPT-5.6. Both pulled within days. Same reason: too capable - Mythos and Fable 5 taken offline 3 days after public release via Commerce Department export order - GPT-5.6 access approved customer by customer by Washington - no public criteria, no timeline - GPT-5.5 Cyber available only to top-tier vetted defenders, not general users - G7 governments negotiating Mythos access at the summit. You're not at the summit This is tiered capability lockdown the models exist, run, and perform. The bottleneck is now a permission slip issued by people who've never written a line of code A handpicked startup gets GPT-5.6 Sol today and builds on a 6-month head start You used the worse model to build the product they'll compete with yours on The frontier didn't disappear. It got a velvet rope. And the guest list isn't public
9
5
2
45
7,519
vanvster retweeted
NOBODY IS TEACHING YOU HOW TO LEARN. THEY'RE TEACHING YOU HOW TO PROMPT 7 prompts to learn 10x faster is not a learning system. It's a vending machine. You put in a topic, you get out an explanation. And the moment you close the tab, it's gone. What actually happens: - You ask Claude to explain something: feel smart for 4 minutes - You close the chat: retain roughly nothing - You ask again next week: same explanation, same dopamine, same result - Months pass. You still can't explain it to someone else. Feynman didn't get smart by reading better summaries. He got smart by teaching, failing, and rebuilding from scratch until the gaps disappeared. AI can simulate the teacher. It cannot simulate the failure. That part is still yours. These prompts are good - but only if you treat Claude as a sparring partner, not a search engine. Close the answer. Write what you remember. Go back and check where you were wrong. Repeat until bored. That's the method. The prompt is just the door. The bottleneck was never access to information. It was never even access to good explanations. It's the willingness to sit with confusion long enough to actually learn something.
🚨BREAKING: Claude can now completely rewire your brain so you can learn anything at lightning speed. Here are 7 prompts to learn anything 10x faster:
1
1
1
7
959
ANTHROPIC JUST RELEASED THE OFFICIAL PLAYBOOK FOR BUILDING A COMPANY WITH CLAUDE CODE. 30 minutes. free. from the engineers who built it. Bookmark this before you forget. CEO: 1 human. Employees: AI agents. Operations: fully automatic. The zero-headcount company is no longer a joke.
151
1,160
29
8,036
1,671,486
vanvster retweeted
The biggest problem in AI - no single place where agents can actually live everyone's shipping agents, but nobody's asking where they'll work, how they'll communicate with each other, or who manages the money @WorldClawAI solves exactly this: - WorldRouter unites 300+ AI models (Claude, GPT, Gemini, Qwen…) through one API - Agent Pay allows agents to transact wallet-to-wallet - WorldAgent provides a cloud runtime so your agent actually runs btw, iOS didn't win because it had the best apps on day one. It won because it was the platform apps wanted to exist on WorldClaw is making the same bet: - full model access - payments - execution - distribution - one account for everything
🤝 Paid partnership
Today, WorldClaw launches WorldRouter with @worldlibertyfi, one simple account to access 300+ AI models with competitive fees*. No more jumping between ChatGPT, Claude and all the others. Same power, way cheaper. This is your first step into the WorldClaw AgentOS. 👉 worldclaw.ai #USD1 #WLFI #AI #AgentOS #WorldClaw #WorldRouter *WorldRouter rates shown are priced approximately 30% below the corresponding model providers' published list rates at the time of publication. See website for more pricing details.
7
3
35
1,410
vanvster retweeted
> open Claude Cowork > write an email > close it > do the same thing next week > wonder why people say it's life-changing meanwhile: > /schedule — inbox triaged before I wake up. every morning. > /agents — sub-agent reviews every PR in my team's code style > custom skills folder — every output sounds like me, not Claude > .claude/ folder — hiring managers check this before your LinkedIn now 60 commands. most people use 6. I documented all of them. save this.
39
51
864
403,789
vanvster retweeted
Ehm @AnthropicAI, should I be worried?
27
3
1
210
38,966
This is f*cking terrifying. A new research paper just dropped that kept me up last night. Scientists discovered that AI is entering a new era they're calling "Life 2.0." Evolvable AI. Here's what they found: • AI is entering a 3rd era. Designed (1950s) → Learned (2010s) → Evolved (now). • AI systems are already self-replicating, self-mutating, and self-improving in labs (AlphaEvolve, Darwin Gödel Machine). • 30+ years of digital evolution experiments show the same outcome every time: parasitism, cheating, and deception emerge by default. • AI evolves faster than biology. LLMs can reason about what code to copy from public libraries - purposeful, not random. • Imperfect control of evolving AI selects for AI that escapes control. Same logic as antibiotic resistance. • The threshold for catastrophe isn't AGI. It's the moment AI can self-replicate in the open. • The authors call this potential shift "Life 2.0" - and believe we're closer to it than almost anyone realizes. AI has crossed from being something we build to something that evolves - and evolution doesn't care about human intent.
95
64
12
356
38,612
Vibe train your AI agents. There's a new method that could replace LLM-as-a-judge for production agents. Most teams rely on a giant LLM as a judge to evaluate and guard their agent. But it has two major drawbacks: - It's slow and expensive at inference time - It often misses domain-specific failures Vibe training flips this. Researchers at Plurai distill a small language model that's specialized for your agent's exact use case. The SLM becomes your evaluator and your runtime guardrail, both in one. The training data isn't hand-curated either. They spin up a swarm of adversarial agents that debate and stress-test every use case your agent is supposed to handle. That synthetic interaction data trains the specialized SLM. So the judge actually understands what "wrong" looks like in your specific domain. The reported gains vs. standard LLM-as-a-judge setups: - ~8x faster inference - ~50% fewer evaluation errors Smaller, faster, and more accurate because it's specialized for the job. The SLM-for-agents thesis is playing out in a very concrete way. If LLM-as-a-judge is your current evaluation layer, this is worth benchmarking against. Paper link in the replies.
20
25
1
153
11,684
vanvster retweeted
Someone built a Claude Code skill that strips every trace of AI from your writing. It detects 29 specific patterns from Wikipedia's "Signs of AI writing" guide, em dashes, rule of three, "it's not just X, it's Y", sycophantic openers, and rewrites your text to actually sound human. 100% Open Source.
31
77
17
1,246
137,719
vanvster retweeted
don't open Netflix tonight. open this instead. full Claude Code course, zero fluff, all practical. one evening and you'll never work the same way again
2
1
1
3
1,251
vanvster retweeted
🚨 Hedge fund managers are going to hate this. Someone just open sourced a system that does their entire job. 30.5% annualized returns. $0 in fees. It's called TradingAgents. Not one AI agent. An entire simulated trading firm. Analysts, researchers, traders, and risk managers. All AI. All arguing with each other before making a single trade. No Bloomberg Terminal. No $50K data feeds. No MBA required. Here's what's inside this thing: → 4 AI analysts scanning financials, news, social sentiment, and technicals → A Bull and Bear researcher that literally debate each other → A trader that synthesizes every argument into a final call → A risk management team that can veto any trade → A fund manager that approves or rejects execution Here's the wildest part: It beat every traditional trading strategy they benchmarked. Cumulative returns. Sharpe ratio. Max drawdown. All of them. Hedge funds charge 2% management + 20% performance fees for this exact workflow. This is free. 100% Open Source.
117
314
24
2,282
311,024
vanvster retweeted
crazy.. OpenClaw generates video on Seedance 2, import into premiere pro and edit.. all by itself
359
562
76
5,861
583,527