Data scientist, inventor and founder.
Joined March 2007
- Tweets648
- Following447
- Followers129
- Likes138
Pinned Tweet
Have idea, will podcast.
I've said for years that any repetitive cognitive task will be taken over by AI.
Tasks in a podcast? Recording words. People saying the words. What the words are.
But the idea, that's all you need now.
Link below: Curiosity Crusaders.
Oracle's AI stock drop shows investors adjusting timelines, not abandoning tech. We're seeing hype evolve to implementation reality. Winners will create advantages applications that solve real business problems with clear ROI paths.
Andrew Duchon retweeted
We replaced a $750K/year marketing team with 12 AI Creator Agents.
No actors.
No products in hand.
No ghost creators.
No missed deadlines.
Just viral TikTok Shop sales — 24/7.
And here’s the wild part:
This system produces 100+ cinematic, product-ready TikTok Shop ads per day from a single prompt.
It feels exactly like running Facebook Ads in 2008…
except CPMs are even lower and everything is organic.
⸻
How the 12-Agent AI System Works 👇
Each Agent is a fully autonomous growth unit that runs its own TikTok Shop profile and handles a complete function:
1. Research Agent
• Trend mining with Kalodata
• Find angles that are already working
• Pull top-performing affiliate hooks
2. Cloning Agent
• Paste a competitor ad → pick an avatar → regenerate
• Auto-rewrite scripts + recreate ads with Sora 2 Pro or Veo3.1
• Produces near-instant product variations
3. Outreach Agent
• Automated creator outreach using Euka AI
• Books creators, negotiates, and distributes briefs
• No manual DMs ever again
4–7. Content Generation Agents
• Daily content produced with:
– Sora 2 Pro
– Veo3.1
– HeyGen
– RunwayML
– Hedra
– Kling
• 10+ video formats per product
• Multiple avatars and voice styles
• Hooks, POVs, testimonials, chaotic edits — automatically
8. Localization Agent
• Auto-transforms videos into 6–10 new markets
• Captions, language swaps, subtitles, voiceovers
9. Repurposing Agent
• Converts a single video into 20+ variations (angles, lengths, CTAs)
10. Compliance Agent
• Removes banned phrases
• Fixes overlays, captions, accuracy claims
• Ensures TikTok Shop approval instantly
11. Optimization Agent
• Iterates based on engagement + CTR
• Swaps hooks, structures, backgrounds
• A/B tests hundreds of versions at machine speed
12. Posting Swarm Agent
• Auto-posts across hundreds of AI creator profiles
• Rotates formats to avoid fatigue
• Drives nonstop organic traffic to your Shop
⸻
No touchpoints. No bottlenecks. No limits.
Just a decentralized AI and UGC force that sells while you sleep.
⸻
Real results from the brands using this:
• As low as $0.10 CPMs
• Thousands of organic views daily
• $69k+ GMV in a month for brands that we helped to launch
• up to 100–300 product-ready videos per day
• $5–8 cost per video (vs. $50–$150 with humans)
⸻
This is the Creator Agent Method.
A plug-and-play system that replaces creative teams and scales content at a speed humans simply can’t match.
I packaged all 12 Agents + the AI V2 workflow so you can deploy the exact system for your brand.
Comment “AGENT” and I’ll DM you everything for free.
(Deleting soon)
P.S. Repost for early access to the full 12-Agent Stack.
Andrew Duchon retweeted
wan 2.2 is turning one attention-grabbing clip into an entire ugc engine
same product, same idea
but dozens of different videos with new faces, pacing, and reactions every time
each clip feels native to the feed, not “ai-generated”
that’s why these fake tip / diy / routine videos are blowing up right now
you feed it one reference and let the model remix endlessly
the algorithm tests everything, keeps the winners, drops the rest
this is how people are flooding tiktok with content without creators or filming
i put together a breakdown on how we run this with wan 2.2
rt + comment “wan” and i’ll send it
(follow for dm)
AI is reshaping how we connect and work. For advisors, success means using AI strategically—automating tasks while enhancing client relationships. I help you implement AI where it adds value, giving you more time for meaningful client connections.
Andrew Duchon retweeted
HOLY SH*T… This AI Agent does everything 🤯
Built in n8n :
🔁 Clones viral TikToks
✍️ Rewrites w/ GPT-4o
🎥 Auto creates avatar videos
🎬 Adds captions & edits
📤 Posts to 9 platforms (TikTok, IG, YT, X…)
🔁 Like+RT
✅ Reply “Steal”
🤝 Follow me & I’ll DM you workflow FREE.
Andrew Duchon retweeted
280k a month in sales from ai videos built off one product prompt
no cameras
no actors
no editors
ai creates dozens of clips that post automatically on tiktok and instagram
each video hits a different angle — authority, curiosity, emotion — all driving to one product
traffic runs 24/7
sales come in daily
i made a short guide showing how the system works
rt + comment "ads" and i’ll send it
(must follow for dm)
Andrew Duchon retweeted
I make $3K a day selling UGC ads.
But here’s what surprises people…
- UGC ads in seconds
- 20x reach in 24/7
- Zero agencies, zero waiting
Every video I sell costs under $5 to produce —
and brands pay hundreds because they perform.
This is the biggest shift in digital ad creation right now.
Comment + RT “UGC” and I’ll DM you exactly how it works.
(must be following)
AI bubble? Maybe - high valuations vs capabilities. But under the hype lies real transformation with current productivity gains. Winners solve specific problems thoughtfully, not adopt blindly. Market correction likely, but core advances continue.
Andrew Duchon retweeted
New Llama.cpp UI is a blessing for the local AI world 🌎
- Blazing fast, beautiful, and private (ofc)
- Use 150,000+ GGUF models in a super slick UI
- Drop in PDFs, images, or text documents
- Branch and edit conversations anytime
- Parallel chats and image processing
- Math and code rendering
- Constrained generation with JSON schema supported
Easy setup + open-source + community-built 🔥
Andrew Duchon retweeted
I spent months illustrating how Transformers actually work.
Not just what they do, but why they’re built this way. The history, design choices, and intuition behind every layer.
From RNNs → Attention → Multi-Head → FFNs → Positional Encoding.
Here's everything I wish I knew:
Andrew Duchon retweeted
Knowledge graphs are infinitely better than vector search for building the memory of AI agents.
With five lines of code, you can build a knowledge graph with your data.
When you see the results, you'll never go back to vector-mediocrity-land.
Here is a quick video:
Andrew Duchon retweeted
I'm blown away by Claude Code powered by Claude Opus 4. I can hand it entire feature requests and let it run for an hour+ in the background.
These new models are now live to Claude Code users. Run `/model` to select the model. Get started here -
claude.ai/code 🎉
Andrew Duchon retweeted
This is Demis Hassabis.
He’s Google's DeepMind CEO, who’s studied AI for 30+ years.
In his new interview with 60 Minutes, he revealed mind-blowing facts about health & AI that 99% of people wouldn’t know.
Here are my top 8 takeaways: 🧵
(No. 6 will shock you)
cannot wait to check this out!
Wow. This really could change everything.
A couple reflections on the quantum computing breakthrough we just announced...
Most of us grew up learning there are three main types of matter that matter: solid, liquid, and gas. Today, that changed.
After a nearly 20 year pursuit, we’ve created an entirely new state of matter, unlocked by a new class of materials, topoconductors, that enable a fundamental leap in computing.
It powers Majorana 1, the first quantum processing unit built on a topological core.
We believe this breakthrough will allow us to create a truly meaningful quantum computer not in decades, as some have predicted, but in years.
The qubits created with topoconductors are faster, more reliable, and smaller.
They are 1/100th of a millimeter, meaning we now have a clear path to a million-qubit processor.
Imagine a chip that can fit in the palm of your hand yet is capable of solving problems that even all the computers on Earth today combined could not!
Sometimes researchers have to work on things for decades to make progress possible.
It takes patience and persistence to have big impact in the world.
And I am glad we get the opportunity to do just that at Microsoft.
This is our focus: When productivity rises, economies grow faster, benefiting every sector and every corner of the globe.
It’s not about hyping tech; it’s about building technology that truly serves the world.
Readers added context they thought people might want to know
Microsoft's supporting paper, published in Nature, does not support the claim that they have created a topological qubit. nature.com/articles/s4158…
Peer reviewers of the Nature paper expressed concern that the paper misleadingly implies that a topological qubit was demonstrated or otherwise achieved: static-content.springer.com/esm/art%3A10.1…
Andrew Duchon retweeted
Announcing How Transformer LLMs Work, created with @JayAlammar and @MaartenGr, co-authors of the beautifully illustrated book, “Hands-On Large Language Models.”
This course offers a deep dive into the inner workings of the transformer architecture that powers large language models (LLMs).
The transformer architecture revolutionized generative AI; in fact, the "GPT" in ChatGPT stands for "Generative Pre-Trained Transformer." Originally introduced in the Google Brain team's groundbreaking 2017 paper "Attention Is All You Need," by Vaswani and others, transformers were a highly scalable model for machine translation tasks. Variants of this architecture now power today’s LLMs such as those from OpenAI, Google, Meta, Cohere, Anthropic and DeepSeek.
In this course, you’ll learn in detail how LLMs process text. You'll also work through code examples that illustrate that transformer's individual components.
In details, you’ll learn:
- How the representation of language has evolved, from Bag-of-Words to Word2Vec embeddings to the transformer architecture that captures a word's meanings taking into account the context of other words in the input.
- How inputs are broken down into tokens before they are sent to the language model.
- The details of a transformer's main stages: Tokenization and embedding, the stack of transformer blocks, and the language model head.
- The inner workings of the transformer block, including attention, which calculates relevance scores, and the feedforward layer, which incorporates stored information learned in training.
- How cached calculations make transformers faster.
- Some of the most recent ideas in the latest models such as Mixture-of-Experts (MoE) which uses multiple sub-models and a router on each layer to improve the quality of LLMs.
By the end of this course, you’ll have a deep understanding of how LLMs actually process text and be able to read through papers describing the latest models and understand the details.
Gaining this intuition will improve your approach to building LLM applications.
Please sign up here: deeplearning.ai/short-course…
Andrew Duchon retweeted
teapawt.net/vitrupo/status/1887475… "AI is better than any Indian programmer that's outsourced right now." Emad Mostaque warns of the "complete destruction" of the Business Process Outsourcing (BPO) market due to AI in 2025. This is the last year when Human work will matter.