This post was originally published on this site
In less than a week, DeepSeek R1, a new AI model from Chinese company DeepSeek, shot to the top of the app stores and rattled Wall Street. Why? This open-source AI tool rivals, and in some areas surpasses, industry leader OpenAI’s capabilities — and at a fraction of the cost. DeepSeek claims it spent less than an estimated $6 million over two months to develop R1.
Two key innovations set DeepSeek R1 apart:
Advanced Reinforcement Learning: Unlike other models that rely on human feedback for refinement, DeepSeek R1 uses a technique called “reinforcement learning” without human input. This allows the model to make decisions and achieve goals more independently. Efficient Distillation: DeepSeek employs an advanced distillation technique, enabling smaller models to extract information from larger ones while using less memory and computing power. This contributes significantly to its cost-effectiveness and rapid development.
Technically, R1 is very impressive, especially considering it was made without access to advanced Western AI chips. The U.S. government’s AI chip export restrictions are clearly driving Chinese companies to next-level creative efficiency. However, despite its achievements, DeepSeek R1 comes with serious concerns around trust, propaganda and censorship. Also at play here, China’s communist government maintains close ties with its large tech companies, including DeepSeek. This relationship, similar to concerns raised about TikTok, adds another layer of complexity when considering the use of DeepSeek R1 in Western newsrooms.
For local newsrooms, DeepSeek presents a complex challenge: How to harness its potential while navigating significant journalism-related risks?
DeepSeek R1’s Potential Newsroom Benefits
DeepSeek R1 offers