A new Google model is nearly perfect on automated handwriting recognition (generativehistory.substack.com)

Article summary

A new Google model, potentially Gemini-3, has shown impressive results in automated handwriting recognition, achieving near-perfect transcription of historical documents. The model's performance is significantly better than its predecessors, with a character error rate of 1.7% and word error rate of 6.5%. The model's ability to recognize handwriting has implications for historians and researchers who work with historical documents. The model's capabilities also raise questions about the potential for AI to make breakthroughs in various fields.

Main themes

• Handwriting recognition
• AI capabilities
• Historical document analysis
• Machine learning
• Innovation and breakthroughs

What commenters say

• The new Google model's ability to recognize handwriting is a significant achievement, but its potential to make novel contributions to various fields is still uncertain.
• Some argue that current AI systems, including the new Google model, are not yet capable of producing truly novel or superhuman intelligence.
• Others believe that AI systems can create novel instances of existing systems or features, but may not be able to expand beyond that into entirely new areas.
• The definition of 'novel' is unclear in the context of AI, making it difficult to determine whether AI systems are capable of producing truly original work.
• The potential for AI to make breakthroughs in various fields is hindered by its reliance on existing knowledge and data, rather than being able to think outside the box.
• Some commentators argue that the idea of AI surpassing human intelligence is overstated, and that current AI systems are not yet capable of replacing human experts in most fields.
• The ability of AI systems to recognize patterns and make connections between existing ideas is impressive, but may not be enough to lead to revolutionary breakthroughs.
• The question of whether AI systems can be creative or truly innovative is still an open one, with some arguing that they are limited by their programming and training data.