{"id":9721,"date":"2025-01-28T20:30:52","date_gmt":"2025-01-29T01:30:52","guid":{"rendered":"https:\/\/www.inthacity.com\/blog\/uncategorized\/deepseek-crushed-big-tech-janus-pro-shocking-ai-model\/"},"modified":"2025-01-28T23:37:00","modified_gmt":"2025-01-29T04:37:00","slug":"deepseek-crushed-big-tech-janus-pro-shocking-ai-model","status":"publish","type":"post","link":"https:\/\/www.inthacity.com\/blog\/tech\/ai\/deepseek-crushed-big-tech-janus-pro-shocking-ai-model\/","title":{"rendered":"DeepSeek Just CRUSHED Big Tech AGAIN With JANUS PRO – New SHOCKING AI Model!"},"content":{"rendered":"

What if I told you that a relatively unknown AI company from Hong Kong, China, is making waves in the tech world by outperforming giants like OpenAI and Pixar Alpha? That\u2019s exactly what DeepSeek is doing with its groundbreaking AI models, Janice Pro and R1. This article dives into the latest video from the AI Revolution YouTube Channel<\/a> to unpack how DeepSeek is redefining the AI landscape. Buckle up, because this story is equal parts inspiring, controversial, and downright wild.<\/p>\n

DeepSeek has been grabbing headlines for two major reasons. First, they\u2019ve released a new multimodal AI model family called Janice Pro, which supposedly beats OpenAI\u2019s DALL-E 3, Pixar Alpha, and Emu3 Gen on benchmarks like Gen-Eval and DPG Bench. These benchmarks measure how well AI models perform tasks like image generation and understanding. The kicker? DeepSeek\u2019s flagship model, Janice Pro 7B, seems to outperform these well-known models\u2014at least according to DeepSeek\u2019s own internal tests. But that\u2019s just the beginning of the story.<\/p>\n

The Rise of Janice Pro: A Multimodal Marvel<\/h2>\n

Janice Pro is a unified Transformer architecture that can handle everything from image generation at 768 x 768 resolution to image analysis and text-based tasks. This all-in-one approach is a big deal because most AI models specialize in just one area, like text or image generation. Think of it as the Swiss Army knife of AI. It\u2019s similar to GPT-4 Vision<\/a>, but with one crucial difference: Janice Pro is completely open source. DeepSeek has released the model\u2019s code and weights on Hugging Face<\/a>, making it accessible to anyone who wants to tinker with it.<\/p>\n

But how good is it really? Users have tested Janice Pro in various ways, from analyzing images to generating scenes. For example, one test involved creating a \u201ccute baby fox in an Autumn scene.\u201d Janice Pro nailed the \u201cbaby\u201d part better than Stable Diffusion<\/a>, but the latter produced a crisper, more polished image. The trade-off? Janice Pro was more faithful to the prompt, while Stable Diffusion excelled in visual finesse.<\/p>\n

DeepSeek\u2019s R1 Model: Matching GPT-4 at a Fraction of the Cost<\/h2>\n

Just days before releasing Janice Pro, DeepSeek made headlines with its R1 language model. This model reportedly matches the performance of GPT-4 but was developed for just $5-6 million. Compare that to the billions spent by Silicon Valley giants like OpenAI and Meta<\/a>. This raises a critical question: Are we overpaying for AI development? Could the next big breakthroughs come from smaller, more agile teams with fresh ideas?<\/p>\n

DeepSeek\u2019s success is particularly intriguing given the geopolitical context. The U.S. has imposed export controls on advanced chips, particularly from NVIDIA<\/a>, to slow down Chinese AI progress. Yet, DeepSeek claims it achieved GPT-4-like results using NVIDIA\u2019s H800 chips, which are technically less powerful than the high-end chips blocked by the U.S. This calls into question the effectiveness of these export controls and highlights DeepSeek\u2019s resourcefulness.<\/p>\n\t\t\t

\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t\t
\n\t\t\t\t\t\t\t\t\t\t\t\t
\n\t\t\t\t\t\t\t
<\/div>\n\t\t\t\t\t\t<\/div>\n\t\t\t\t\t\t
<\/div>\n\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\n\t\t\t\t\t
\nPrev<\/a>\n1<\/span> of<\/span> 1<\/span>\nNext<\/a>\n<\/div>\n
\n\t