How to Use DeepSeek

About this video

This video gives a practical look at DeepSeek amid hype that the open-source model from China rivals ChatGPT and Claude, showing how to use most of it in five minutes and comparing quality, cost and privacy. Sign-up requires an account; the presenter recommends email and password over connecting Google for privacy, warns a personal email must be used because a business email will be rejected, and describes entering a code sent by email. He turns on DeepThink, a reasoning mode that exposes the chain of thought, takes extra time — about 15 seconds for a letter-counting question ChatGPT has stumbled on — and checks itself repeatedly for a more accurate answer. Examples include a fence-painting work-rate problem with Alice painting in four hours and Bob in three, and a complex 1954 smallpox-eradication planning prompt, while a simple twist shows DeepThink is not needed for easy questions. In side-by-side tests against ChatGPT’s comparable reasoning model, both produce solid productivity and fundraising answers, with DeepSeek more tactical and ChatGPT more strategic and faster — six seconds versus DeepSeek’s 54 on one test; ChatGPT’s two-player Pong game is better though DeepSeek can run HTML inline, DeepSeek’s haiku is judged more cohesive, and DeepSeek search fails in the demo while ChatGPT search works. On cost he cites reported training costs of $60 million for Meta’s Llama 3.1, $100 million for OpenAI’s model, and only $5.6 million for DeepSeek V3, and much cheaper token costs: ChatGPT at $20 a month with 50 messages per week for its reasoning model or $200 a month for near-unlimited use, versus no way to pay on DeepSeek’s site and an estimated equivalent around $7. He closes on privacy: the web interface is run out of China, there is no opt-out of data use for training, and running the open-source version locally is presented as the theoretical solution.

Key takeaways

• Sign up with a personal email and password and a emailed code; a business email will be rejected
• Turn on DeepThink for complex problems to get visible chain-of-thought reasoning, but skip it for simple questions to save time
• In the presenter’s comparisons ChatGPT was generally faster and stronger on most tasks, while DeepSeek won the haiku round and can run HTML inline
• The claimed breakthrough is cost and open source: $5.6 million cited for DeepSeek V3 training versus $60 million and $100 million for the Western models cited, with far cheaper token/subscription-equivalent costs
• Privacy is the trade-off: the web service is run out of China with no opt-out of training use, so running the open-source model locally is the suggested alternative

Episode details: published January 31, 2025 · 11:26 long. Summary written by Self Growth Videos from the episode itself. Business results and any income figures mentioned are Kevin Stratvert’s own claims, shared for education — they are not typical or guaranteed.

← Previous: How to Make an Automated Data Entry Form in Excel (Step-by-Step Tutorial) · Next: How to Install DeepSeek on Windows →

All Kevin Stratvert videos · Kevin Stratvert profile · Kevin Stratvert on YouTube