ChatGPT co-inventor unveils Jev, a new kind of frontier AI
What's the story
TypeSafe AI, a start-up led by ChatGPT co-inventor Diogo Almeida, has launched an innovative AI model called Jev. Unlike conventional models that respond to text-based prompts with natural language, Jev is designed for interaction with machines. It provides typed probabilistic decisions for other software or AI models to use. The approach is similar to type safety in programming where errors from unexpected data types are caught during processing.
Model efficiency
Jev is built on an RLCD architecture
Jev is a System One model, built on an architecture called Reinforcement Learning for Calibrated Decisions (RLCD).
The model's efficiency lies in its ability to provide responses much faster than traditional LLMs.
While LLMs have variable response times that can stretch into minutes, Jev's response time ranges from 70ms-500ms, 40x-200x faster than conventional models.
Economic advantage
Jev is also cheaper than traditional LLMs
Jev is also more cost-effective than its competitors. It charges $0.042/MTok for input and nothing for output, while OpenAI's GPT-5.6 Terra costs $2.00/MTok and $12/MTok for output.
The name "Jev" is a reference to 19th-century economist William Stanley Jevons, who argued that efficiency gains from technology can lead to increased consumption of resources like coal rather than reductions in usage.
Use cases
Potential applications of Jev
Jev is well-suited for AI automation software, real-time applications requiring quick decisions, AI map reduce jobs for classifying large data sets, verification of AI inputs, and harnessing other AI models.
However, it's important to note that while TypeSafe claims Jev is hallucination-free (a common issue with LLMs), this isn't a fair comparison as its output isn't natural language but structured responses with probabilities.