Blog to Video Knowledge Product | Generated by AI
Question: Turn jekyll-ai-blog into a video-first knowledge product — a “Geek + Gary Vee + 稚晖君 + BPS Space” style channel. Is this a good idea?
Answer:
I think this idea is very aligned with your current trajectory. The key insight is:
Your blog is already a knowledge production engine. The next evolution is not “make videos randomly”; it is to build a multimodal knowledge distribution system.
Your existing repo lzwjava/jekyll-ai-blog is essentially:
Research → Notes → Markdown → Static Website → Readers
The video version becomes:
Research
↓
AI agent summarizes + structures
↓
Script
↓
Voice/avatar/screen recording
↓
Short video
↓
YouTube / Bilibili / TikTok / WeChat Channels
↓
Backlinks → blog → community → business
The important part: the source of truth remains the blog. Video is just another rendering target.
Positioning
The closest mental model:
Geek + Gary Vee
Gary Vee:
- high frequency
- personal brand
- opinion + distribution
Geek:
- curiosity
- engineering depth
- “I built/tested this”
稚晖君:
- maker credibility
- showing experiments
BPS Space:
- explain frontier technology visually
Your unique angle:
“An AI engineer building and explaining the future in public.”
Not a news commentator.
More like:
I read papers
I run models
I buy weird GPUs
I debug hardware
I build agents
I explain what I learn
Your MI50 experiments, GPT-2 training on RTX 4070, nanoGPT experiments are actually very suitable content.
Content pillars
I would avoid only “AI news”.
AI news is a commodity.
Better:
1. Build in public
Example:
“Can a $680 AMD MI50 train an LLM?”
Video:
0:00 I bought a $100 GPU nobody wants
0:30 Why this GPU was a datacenter monster
2:00 Hardware teardown
4:00 Installing ROCm
7:00 Training GPT-2
10:00 Result
This is very稚晖君-like.
2. Explain frontier AI
Example:
“Why DeepSeek changed AI economics”
Structure:
Problem
↓
Old approach
↓
New idea
↓
Code demonstration
↓
Business impact
3. AI engineer life
Example:
“How I trained GPT-2 from zero on one GPU”
This has authenticity because you actually did it.
Many creators explain things they never touched.
4. AI tools for normal people
This is Gary Vee side.
Example:
“10 minutes: build your own AI employee”
Technical architecture
Your blog can become the content database.
Something like:
posts/
2026-07-mi50-gpt2.md
|
v
AI Pipeline
extract:
- title
- hook
- story
- technical explanation
- diagrams
|
v
video_script.md
|
+---- YouTube long video
|
+---- Shorts
|
+---- Bilibili
|
+---- Blog
Basically:
Jekyll → LLM content compiler → video renderer
AI-native video factory
A simple MVP:
article.md
↓
Claude/GPT
↓
script.json
{
hook:"",
scenes:[
{
narration:"",
visual:"",
code:""
}
]
}
↓
Remotion
↓
mp4
Use:
- Remotion for programmatic video
- ElevenLabs/OpenAI voice
- Whisper for captions
- Mermaid/Manim for diagrams
- ffmpeg pipeline
The advantage:
You can produce 3-5 videos/week without a traditional video team.
But don’t copy Gary Vee too much
Your advantage is not energy.
Your advantage is:
depth × authenticity × AI acceleration
There are already thousands of energetic AI influencers.
Much fewer people can say:
“I trained GPT-2 myself yesterday. Here is the loss curve.”
That is your moat.
Possible brand
Something like:
AI Engineer Diary
or
Building The Future
or
Geek Lab
or
AI Native Builder
The content feeling:
Today I bought a $1000 accelerator.
Let's see if it can train a language model.
No slides.
No marketing.
Just experiment.
I think the strongest strategy is:
Do not create a video channel first. Create a “knowledge compiler”.
Your blog is the database.
Video, newsletter, Twitter/X, YouTube, Bilibili are different frontends.
You are basically building an AI-era version of a technical media company.
The timing is good because AI makes the production cost collapse.
Your old advantage:
one engineer writes one article
New advantage:
one engineer creates:
- article
- video
- podcast
- course
- community
This is exactly the kind of leverage that fits someone who already builds agents and infrastructure.