This week, just one AI news story was enough to dominate the entire week, and perhaps the entire year?
Weekly, I sift through the AI buzz on Fridays. I spotlight what truly matters in AI-fuelled creativity. Explore the week’s standout innovations, carefully ranked for their impact. Stay one step ahead, unleashing your creativity like never before.
AI-power for digital ART is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber. DeepSeek-V3 is a powerful new AI model released on December 26, 2024, representing a significant advancement in open-source AI technology. It is basically the Chinese version of Open AI. They went the same open-source route as Meta. 🟣 DeepSeek outshines 4o on nearly every benchmark, all at just 10% of the cost.
The model features impressive technical capabilities:
- 685 billion total parameters with 37 billion activated parameters
- Trained on 14.8 trillion high-quality tokens
- Processing speed of 60 tokens per second, 3x faster than its predecessor
Training cost of $5.5 million using 2,788,000 H800 GPU hours DeepSeek-V3 demonstrates exceptional performance across multiple domains:
- Outperforms Llama 3.1 405B and GPT-4o in coding competitions on Codeforces
- Shows comparable benchmarks to Claude 3.5 Sonnet
- Excels in integrating new code with existing codebases
Starting February 8th, the model will be available at competitive rates:
- Input: $0.27 per million tokens ($0.07 with cache hits)
- Output: $1.10 per million tokens
The model has some notable restrictions:
- Requires substantial computational resources for unoptimized versions
- Content filtering for certain political topics due to Chinese regulatory requirements. 🟣 has less censorship than Qwen
DeepSeek-V3’s release has influenced the AI market significantly, forcing competitors like ByteDance, Baidu, and Alibaba to reduce their pricing models and offer some services for free. The model’s development by a Chinese company backed by High-Flyer Capital Management demonstrates growing competition in the global AI landscape.
DeepSeek-V3 represents a leap forward in open-source AI, offering high performance at a competitive cost, making it a significant player in the ongoing 2025 evolution of large language models.
https://www.deepseek.comhttps://huggingface.co/spaces/akhaliq/anychat
I’m half thrilled, half nervous about how it might shake up our entire creative process. I’m honestly blown away—this latest development completely caught me off guard and shifted everything I thought I knew. Some are calling it the biggest shake-up of the entire year. The sheer scale of it leaves me both thrilled and a touch unsettled.
