Search
DeepSeek has just rolled out a revolutionary Optical Character Recognition (OCR) model, leveraging the timeless wisdom that "a picture is worth a thousand words" to surmount the computational bottlenecks plaguing AI when it comes to processing lengthy documents. This innovative model, boasting a mere 3 billion parameters, dramatically boosts processing efficiency by converting text into image format. Remarkably, it can churn out 200,000 pages of training data daily with just a single graphics card. This feat has garnered accolades from industry experts such as Karpathy, who have likened it to the "JPEG moment for AI"—a nod to the transformative impact JPEG had on image compression. Moreover, there's been speculation within the industry that this model might inadvertently reveal some of Google Gemini's core technical secrets.
-
C114 Communication Network -
Communication Home
