My view on DeepSeek (4/N) N=4 Back in 2023, DeepSeek was still standing at the stage's edges. By 2025, right before Chinese New Year, it has risen to the pinnacle of the AI world.   4.Argo and DeepSeek   As a developer at Argo and a researcher in AIGC, I’ve DeepSeek-ified key functions within Argo. Argo, a workflow system, uses DeepSeek R1 to generate rough initial workflows. We’ve also embedded DeepSeek R1 as the default LLM, opting out of expensive, closed-source models like OpenAI’s. Since workflow systems typically involve large token consumption and extensive context (on average ≥ 10k tokens), using higher-priced models like OpenAI’s or Claude 3.5 would be prohibitively costly. That kind of preemptive spending can hurt a product before web3 users truly capture any value.   As DeepSeek continues to improve, Argo will deepen its collaboration with the rising “Chinese contingent” led by DeepSeek, including (but not limited to) localizing text-to-image/video interfaces and LLM development for the Chinese market.   In terms of partnerships, Argo plans to invite DeepSeek researchers to share their technical achievements and offer grants for top AI researchers. This will help web3 investors and users stay up-to-date with AI advancements and foster broader community engagement.