sometimes I feel AI smart

From Word2Vec to Char-RNN to GPT-2 — and who knows, perhaps GPT-6 is just around the corner. I've been using ChatGPT for three years now. I started with ChatGPT Plus for coding back in 2022, and from 2023 onward, I've been exploring AI tools for fun, mainly generate AI music.

Today, I want to talk about the intelligence, which depends on big data. more data more smart. A model somelike a huge knowledge base, Reasoning is Search. For example, an agent should be smart enough to build a React web app, but perhaps not as smart when building a Python rich library tool. I have gained several valuable insights from using AI tools.

some moments I feel AI very smart:

  1. when ChatGPT came out, it knew so much
  2. ChatGPT plus as a travel assistant to translate or ask anything in an image when go to some place like Thailand or Vietnam
  3. analyze and infer a news from hn news
  4. agent-browser on openClaw, play youtube MV for me by feishu channel.
  5. openclaw skill-creator. I've created 30 skills with it. clawhub.ai/u/goog
  6. use brower-use.com
  7. hermes Agent-Managed Skills
  8. call openrouter API, it will call web_search auto when needed
  9. use KIMI agent to build web apps
  10. use claude code cli to upgrade or debug my web app
  11. AI music(suno and haimian) sounds good
  12. when I scraped data, openClaw started a subagent for me >I'll scrape all pages from 101 Cookbooks vegetarian recipes systematically. Since direct fetch is blocked, let me use web search page by page to gather recipe data, then append to the existing database.

Large Language Models(LLMs) are good at handling structured data like markdown and json formats, skills and workflows. when one agent has many skills and performs well, it could replace some human work.

some actions to hug AI

  1. Use AI chatgpt more often and rely less on web search
  2. i have AI coding(vibe coding/agent coding) two months. ClawHub coding project demo

Interesting projects

  1. browser-use
  2. agentmail.to
  3. coze.cn
  4. can1357/oh-my-pi: ⌥ AI Coding agent for the terminal — hash-anchored edits, optimized tool harness, LSP, Python, browser, subagents, and more

Time flows fast

I signed up for ChatGPT in 2023.02. After three years, the market share of AI models has changed a lot. ChatGPT doesn’t surprise me anymore, but I’m still waiting for GPT-6. Claude code help me code some projects, LLM models like mimo and minimax help feed the openClaw. In 2025, LLMs have made tremendous strides; Their abilities have grown a lot, and now that they seem to have mastered programming, they can automate tons of tasks. Consequently, many people have lost their jobs—particularly programmers in developing countries, who may find themselves facing severe financial hardship. define levels of GPT 2023 >And you realise you did not choose any of the changes. They happened while you were looking the other way, and now they are your life. -- ifthisroad.com>

No matter you choose or not, some changes by AI have happened.

Some questions to think

Here’s my questions to ask AI model providers and some IA groups:

  • Will AI benefit poor and everyday people?
  • Could AI-powered drones pose a threat to human safety?
  • Should AI providers compensate users when a model gives incorrect or harmful answers?
  • Where is humanity headed in an AI-driven future?