Clef: Open-weight decision models, and new RL fine-tuning platform (blog.cloudflare.com)

Article summary

Cloudflare is releasing two open-source decision models, Clef and Clef-flash, which are designed to make classifications and decisions based on input data. These models are hosted on Workers AI and are compatible with the Jev Decision Index. Clef has a vision encoder, allowing it to classify visual content, and a 64k context window, making it more powerful than other decision models. The models are available for fine-tuning and can be used for various applications, including threat intelligence and customer support.

Main themes

• Decision Models
• AI Innovation
• Model Creation
• Market Competition
• Automation
• Integration
• Calibration
• AI Advancements

What commenters say

• The concept of decision models is not new and has been around for a while, with some models already existing before Jev.
• Decision models can be easily created using existing open-source models and large language models, making them not particularly innovative.
• The ease of creating decision models means that many companies and individuals can create their own models, potentially leading to a flooded market.
• The value of decision models lies in their ability to make consistent and predictable decisions, with near 100% determinism, which is useful for automation.
• The pricing of decision models, such as Clef, is competitive with other models like Jev, but may not be the most important factor in choosing a model.
• The integration of decision models with other systems and tools is more important than the models themselves in determining their usefulness and market capture.
• Calibrating decision models to produce meaningful probabilities is important, but may not be a major differentiator between models.
• The field of AI is advancing rapidly, and decision models are just one example of the many unexplored avenues and low-hanging fruits that exist.