TypeSafe AI has launched Jev, a transformer-based model that outputs probabilities instead of text, reducing automation costs by up to 95% according to developers.
TL;DR
- TypeSafe AI introduces Jev, a model designed for automation that outputs probabilities, not text.
- Developers report significant cost savings and speed improvements compared to traditional LLMs.
- Jev's unique architecture and synthetic data training set it apart from conventional AI models.
What happened
Diogo Almeida, a former OpenAI researcher, founded TypeSafe AI to address the limitations of large language models (LLMs).
This week, TypeSafe AI released Jev, a transformer-based model that produces calibrated decisions instead of text outputs.
Jev's design makes it significantly cheaper and faster than traditional LLMs, with input tokens metered by the billion.
The model has garnered significant interest from developers, briefly overwhelming TypeSafe AI's API due to high demand.
Why it matters
Jev's ability to output probabilities makes it ideal for automating workflows, providing real probabilities that other models lack.
Developers have reported speed improvements of five to 18 times and cost savings of up to 95% compared to traditional LLMs.
Jev can be used to augment LLMs, acting as a smart check on misbehavior and preventing jailbreaks.
The model's low cost and speed make real-time sorting of workloads possible, a feature that could be highly valuable for developers.
Key facts
- Jev is a transformer-based model that outputs probabilities, not text.
- Developers report cost savings of up to 95% and speed improvements of five to 18 times compared to traditional LLMs.
- Jev is trained exclusively on synthetic data using a technique called 'reinforcement learning from calibrated decisions'.
- The model is named after William Stanley Jevons, a 19th-century economist known for Jevons' paradox.
- TypeSafe AI briefly lost the ability to serve users from its API due to high demand for Jev.
- Jev's architecture is suspected to be built on top of an open-weight LLM, but the company has not confirmed this.
- TypeSafe AI plans to build more versions of the model in new modalities.
Context
Jev's release comes at a time when the AI industry is increasingly focused on automation and cost efficiency.
The model's unique approach to outputting probabilities sets it apart from traditional LLMs, which are optimized for human language.
TypeSafe AI's focus on synthetic data training is a departure from the industry norm, which typically relies on large datasets of human-generated text.
