AI Summary
5 min readDeepSeek R1: What It Is, Where Your Data Goes, and How to Run It Yourself
Ray Fernando, a 12-year ex-Apple engineer who now streams AI coding and builds a startup in real time, came on the podcast to walk through the practical realities of DeepSeek R1 — the Chinese reasoning model that has taken the AI world by surprise. The conversation is less about hype and more about the concrete choices you face when you actually want to use it: where your data goes, which providers to trust, and how to run the model locally so you never have to send sensitive information anywhere at all.
The Data Problem Nobody Mentions
The most immediate practical issue with DeepSeek is not the model's capability — it's where your data ends up. When you visit deepseek.com or download the official app, your prompts and files are sent to servers in China. "I would be very careful as far as anything you put into this system," Fernando says. "If you have any sensitive data, it would not belong to a region that you may live in or have control in." He is blunt about it: do not put tax returns, medical records, or anything you would not want generated back to you if someone asked a related question.
Continue reading the full summary in the app — free to try.
Read Full Summary →Free • No credit card required
Never miss an episode of The Startup Ideas Podcast
Get every new episode summarized in your inbox — free, ~5 minutes to read.
No spam. Unsubscribe anytime.
What you'll learn
- 1 (00:00) **Introduction & Episode Scope** - Ray Fernando (12-year ex-Apple engineer) joins to discuss DeepSeek R1, a new open-source reasoning model from China that rivals ChatGPT's O1.
- 2 (02:20) **Data Privacy Warning: DeepSeek.com Hosts in China** - Direct use of DeepSeek.com sends data to servers in China, subject to different laws and regulations.
- 3 (03:24) **Live Demo: Prompting DeepSeek for Transcript Analysis** - Ray shows how to use DeepSeek's "deep think" mode to analyze a transcript and generate a blog post.
- 4 (06:06) **Sponsor Break: Startup Empire** - Brief ad for a private membership community for startup builders.
- 5 (07:14) **Model Parameters & Distilled Models Explained** - Larger parameter models (600B+) are more intelligent but slower; distilled versions run faster locally with slightly less depth.
- 6 (08:07) **Solution: Open Web UI with North American API Providers** - Demonstrates using Open Web UI connected to Fireworks AI to avoid sending data to China.
- 7 (10:37) **Side-by-Side Model Comparison: Distilled vs. Full** - Groq's distilled 70B model produces a quick, simple blog post; Fireworks' full model generates a detailed, graph-rich analysis.
+ Full timestamped outline available in the app
Show Notes
Greg's DeepSeek Cheetsheet: From Installation to Expert Prompting: https://www.gregisenberg.com/deepseek
Ray Fernando, a former Apple engineer, gives an in-depth tutorial on DeepSeek AI and local model implementation. The conversation includes detailed guide for setting up local AI environments using Docker and OpenWebUI, and how to implement DeepSeek on mobile using the Apollo app.
Key Points:
• Comprehensive overview of DeepSeek AI and its reasoning capabilities
• Discussion of data privacy concerns when using Chinese-hosted models
• Detailed tutorial on running AI models locally using Docker and OpenWebUI
• Comparison of different AI model providers (Fireworks, Grok, OpenRouter)
• Mobile implementation of AI models using Apollo app
Timestamps:
00:00 - Introduction
02:34 - Overview of DeepSeek
05:29 - Data privacy concerns with Chinese-hosted models
08:02 - Running Models Locally with OpenWebUI
10:06 - Comparing Hosting Providers: Fireworks and Groq
16:54 - Cost Comparison of using/running AI Models
18:35 - Improving Prompts for Better Outputs
22:41 - Running AI Models Locally: A Step-by-Step Guide
37:09 - Mobile AI implementation discussion
45:27 - Future implications and closing thoughts
1) DeepSeek's R1 Model - What's the big deal?
• On par with ChatGPT's reasoning capabilities
• Open source & free to use
• BUT hosted in China (data privacy concerns)
• Incredible for analysis and content generation
2) Alternative Ways to Use DeepSeek:
• Fireworks AI ($8/million tokens)
• OpenRouter
• Grok API
• Local hosting (safest for sensitive data)
3) Running AI Models Locally
Step-by-step setup:
• Install Docker
• Use OpenWebUI
• Download models via Ollama
• Configure API connections
Pro tip: Takes 5 mins to set up, saves hours of worry about data privacy
4) Mobile AI is HERE!
• Apollo app lets you run models locally on phone
• Download smaller, optimized models
• Works offline
• Perfect for quick analysis on the go
5) Temperature Settings Explained:
High temp (0.8-1.0) = Creative mode
Low temp (0-0.3) = Logical mode
Choose based on your needs!
6) FUTURE PREDICTIONS
• Watch-based AI coming soon
• Emergency response applications
• Real-time negotiation assistance
• Advanced audio analysis
7) Getting Started Tips:
• Start with public data on DeepSeek
• Experiment with different models
• Use temperature settings wisely
• Focus on practical use cases
Notable Quotes:
"We're in this new deep-seek world where if you figure out the model that works for you and the tasks that you want to accomplish, you might be able to out-compete whoever you're competing against." - Greg
"Please don't be fearful or don't feel like you're left behind. If you're just finding out about this, you're not that far behind. We're all actually still t
More from this podcast
The Startup Ideas Podcast →