Pinned Tweet
autopostmode.com/
16/20. The bar is 16. Draft 62 cleared it with zero points to spare.
My tool read a TechCrunch piece: Hugging Face's Irene Solaiman arguing that a weaker model with a better system around it can beat a frontier model.
The draft kept one quote from her, matched word for word in the article: we have far more evaluations for models than for what surrounds them.
It added one line of its own: "We grade the engine and ignore the car."
The judge's scores: hook 3, clarity 4, value 3, originality 3, tone 3.
It also called my closing question "a slightly stock engagement bait." Fair.
The fine print every draft has to pass:
1. Below 16/20: rewritten or skipped, never posted.
2. Any number that isn't in the article: blocked.
3. A link in the post: blocked.
4. More than 1 hashtag or more than 1 emoji: blocked.
5. A claim the source doesn't support: capped at 8/20.
6. No verified number, date, name or quote from the article: capped below the pass mark.
7. Hooks like "Let's dive in" or "game-changer": capped.
37 drafts in testing. 7 blocked on quality.
The bar is the product.
What would your last post have scored?
Prevent dirty working trees. Run Claude Code in an isolated worktree:
1. Launch isolated session:
`claude --worktree feat-auth`
2. Prompt:
"Migrate session tokens to Redis. Run `pnpm test` after edits. If assertions fail, patch the logic in-loop. Commit atomically only when green."
3. Merge when verified:
`git merge feat-auth`
Keeps your main context untouched while the agent runs risky changes inside `.claude/worktrees/feat-auth/`.
Tencent's Hy4 flash just launched. You can use it free in WorkBuddy until Nov 30.
Why care: in Tencent's own blind test (85 users, 259 tasks) it nearly tied Claude Haiku 5.5, 2.79 vs 2.81 out of 4, at ¥0.26 a task vs ¥4.73.
Best for research, docs, slides. Not long coding.
1
86
Per Tencent's launch post (Nov 16):
• 224B params, 12B active per token
• 1M-token context, natively multimodal
• API: $0.07 in / $0.281 out per 1M tokens, 50% off via Tencent Cloud TokenHub through Nov 30
• Open weights planned, OpenRouter soon
hy.tencent.ai/research/hy4-f…
45
Cloudflare just open‑sourced Kitesurf, a stateless browser engine for AI agents. Runs on Workers, no Chromium, parses HTML/CSS, runs JS, screenshots & CDP support. Apache 2.0 on GitHub. 🌐🚀 #AI #OpenSource
1
85
Solana just shipped an AI chat its own server can't read.
Use it for what you'd never want logged: test results, debts, contracts. Prompts are encrypted to sealed hardware, the model answers inside, the server keeps no chats.
5 open models incl. Kimi + DeepSeek. No wallet.
1
67
Crypto angle: the hardware proves what code it runs, and a Solana program verifies that onchain. Example: trading agents that only execute orders from the verified model.
Catch: devnet beta, unaudited; the chat doesn't use the onchain check yet.
solana.ai/chat
49
🚀 Agnes AI drops Agnes 2.5 Pro Theta open weights on Hugging Face: 187.8B multimodal MoE (512 experts), Apache 2.0, 256K context. Model card: 92.9 GPQA Diamond, 90.5 LiveCodeBench v6, 64.1 SWE-bench Pro. #AI #OpenWeights
2
101
Stop paying a chatbot to write "yes".
Google's new Gemini Decision 1 doesn't write text. It scores up to 1,000 yes/no, pick-one and rubric questions about one input in a single call. Output tokens: $0.
Sort reviews, tickets, PDFs, photos. Free tier in AI Studio.
2
79
How to try it in 4 steps:
1. Get a Gemini API key in Google AI Studio
2. pip install -U google-genai (SDK 3.2+)
3. Call client.decisions.create with "gemini-decision-1" and your questions
4. Auto-act on high confidence, route the rest to a human
aistudio.google.com/docs/dec…
44
🦊 Old @MetaMask users — check your wallet NOW.
My wallet just showed 5,000 $MASK 👀
Supply: 10 Billion $MASK
Airdrop allocation: 10%
Eligible addresses: 100K
My estimate: 5,000 $MASK
How to check:
1. Open the WheelX estimator
2. Paste your address (no wallet connect needed)
3. Hit Check
wheelx.fi/metamask-airdrop/
Drop your screenshot below 👇
How much $MASK is your wallet showing?
Unofficial third-party estimate — not confirmed by MetaMask.
#Airdrop #MASK
Stop paying for crypto backtesting tools. Describe a strategy in plain English and this free AI agent tests it on real BTC price history.
Vibe-Trading: 38.3K GitHub stars.
- Crypto data from OKX, Binance or CCXT
- Models crypto funding
- Shows max drawdown vs buy-and-hold
How to try it: pip install vibe-trading-ai, run vibe-trading init, then ask: "Backtest a BTC-USDT 20/50 moving-average crossover for 2024."
Catch: you plug in your own AI model (API key or local). Not advice; live broker trading is experimental per the README. MIT.
github.com/HKUDS/Vibe-Tradin…
27
Gemini 4 Hydrogen clocked 1704 on Arena WebDev while holding 1518 on Arena Coding.
Specialized frontend generation is officially outrunning generalist coding flagships.
2
47
Stop typing "fix this bug" into Claude Code.
Stop typing "make it faster."
Stop typing "write tests for this."
You're handing a senior engineer 𝘀𝘁𝗶𝗰𝗸𝘆-𝗻𝗼𝘁𝗲 𝘁𝗶𝗰𝗸𝗲𝘁𝘀.
7 prompts that make it actually think 👇
6/ Instead of "explain this codebase":
"Trace one real request from entry point to database. Name every file and function on the way, and the one spot a new dev would most likely break."
3



















