
The ENTIRE History of Deepseek in 26 Minutes
Title history
- The ENTIRE History of Deepseek in 26 Minutes — observed on 20261114-1727 UTC (YouTube)
Keywords
Summary
171 words
Critical Evaluation
Value of the Information & Strength of the Argument
The video provides a coherent and well-structured narrative, integrating a wealth of factual events with context. Its argumentation is solid, presenting cause-and-effect relationships (e.g., cost efficiency leading to market disruption) backed by specific figures (training costs, market losses, funding amounts). The story is compelling and avoids overt bias, presenting both Western and Chinese perspectives. However, it occasionally speculates on internal motivations (e.g., ‘Liang’s bet on curiosity’) without direct evidence, and the rapid-fire timeline may oversimplify complex technical and geopolitical nuances.
Scientific Rigor, Source Quality, Title Accuracy
The video demonstrates strong rigor by frequently citing established news outlets (Financial Times, Reuters, Bloomberg, South China Morning Post) and the Nature paper. It distinguishes between confirmed reports and leaks (e.g., the July transcript) and acknowledges uncertainties (e.g., ‘reports, not proven findings’). The title is highly adequate, covering the entirety of DeepSeek’s history in a comprehensive yet accessible manner. The use of multiple sources enhances credibility, though the reliance on journalistic accounts rather than primary technical papers limits depth.
174 words
Title / Content Match
The title accurately reflects the content: a comprehensive chronological history of DeepSeek from its origins to late 2026.
Quality & Reliability
7/10
The video synthesizes numerous public reports from reputable outlets such as Financial Times, Reuters, and Bloomberg, and includes peer-reviewed publication in Nature. However, it relies heavily on journalistic sources and unverified leaks, and some claims are presented without direct evidence.
Key Moments
Markers derived by PSI from the transcript: the creator did not define chapters.
- Introduction to Liang Wenfeng's early life and the founding of High-Flyer in 2016.
- Building the Fire-Flyer compute cluster and acquiring 10,000 Nvidia A100 chips.
- DeepSeek officially launches as an AGI lab in May 2023; releases first models (Coder and general-purpose).
- Release of V2 triggers a price war in China, earning DeepSeek the nickname 'Pinduoduo of AI'.
- V3 release on December 26, 2024, with a training cost of only $5.6M, shocking the industry.
- R1 launch and the historic $593 billion Nvidia stock crash on January 27, 2025.
- Government bans and restrictions on DeepSeek across Italy, Taiwan, Australia, and the U.S.
- Liang Wenfeng meets Xi Jinping; the R2 project with Huawei chips fails.
- Nature publishes the R1 paper in September 2025; Anthropic accuses DeepSeek of distillation in February 2026.
- V4 launch, huge funding rounds, and DeepSeek's evolution into a major company.
Cited Sources
- Financial Times — Referenced for reports on DeepSeek's move away from Nvidia and R2 delays.
- Bloomberg Billionaires Index — Cited for Liang Wenfeng's net worth and funding details.
- Reuters — Used for October 2026 funding round report.
- South China Morning Post — Cited for V4 training timeline and 'minor upgrade' reports.
- Nature — Mentioned as the journal that published the R1 paper after peer review.
- Anthropic — Source of the distillation accusation against DeepSeek.
- HuggingFace — Cited for R1's 11 million downloads.
Concurring Sources
- DeepSeek-R1 Paper in Nature — Two peer-reviewed publication confirming R1's reasoning training approach and efficiency.
Dissenting Sources
- Anthropic's Distillation Accusations — Anthropic claimed DeepSeek used unauthorized distillation from its Claude models, which DeepSeek denied.
Contribution & Novelties
The video offers a unique synthesis of DeepSeek’s history, connecting technical milestones with geopolitical and market reactions in a single narrative. It highlights the paradigm shift in AI cost efficiency and the open-source movement, challenging the assumption that massive investment is necessary for frontier AI. The inclusion of recent events (up to late 2026) provides timely context.
Pour aller plus loin :
- DeepSeek on Wikipedia — Overview of the company and its models.
- Reinforcement Learning — The technique used to train R1’s reasoning abilities.
- Sputnik moment — Historical analogy used in the video to describe R1’s impact.
97 words
Radar Profile
The radar shows a balanced profile with high scores in quantity and quality of information, moderate technical depth, and strong overall reliability. The video excels in coverage and sourcing but is less technical, making it accessible to a general audience while retaining credibility.