22:42Machine-translated from Azerbaijani by AIRead in Azerbaijani
Rivals Multiply Rapidly Following TypeSafe’s Jev AI Decision Model
After TypeSafe released a decision model called Jev, the number of similar services and open-source projects in the artificial intelligence market increased rapidly. Conventional chatbots generate text, while these models calculate probability scores that software can use. Nandakishor Mukkunnoth, founder of ConvAI Innovations, says he developed the underlying architecture more than 1 year ago. However, no evidence showing that TypeSafe copied his work has yet been made public.
Jev works as an API. An API is an interface used for software to connect to this service. The system receives a customer’s online purchase information and submitted questions. For example, it can determine whether the buyer wants to return a product and which product the buyer is referring to. The response is provided as a probability score for yes-or-no or multiple-choice options. The developer then uses that number to set conditions in the software. Because the model does not generate text, it does not need previous conversations or extensive context. It processes questions simultaneously and does so at a lower cost than conventional large language models (LLMs). TypeSafe also published a brief technical explanation of the System One architecture. Just a few days later, Mukkunnoth wrote on his blog that he had developed the Jev architecture 1 year earlier. He published a scientific paper about SalesRLAgent, a model that calculates probabilities, on March 30, 2025. He also openly shared the model’s ready-to-use parameter files and dataset. A 2nd study released in September was closer to how the Jev model works. According to Mukkunnoth, SalesRLAgent was overlooked because it only solved a specific problem in sales. TypeSafe turned the same approach into a decision engine that can work across different fields. The finished product includes the company’s servers, API access, documentation, developer tools, and payment plans. Mukkunnoth later released the Laya model as open source. Users can run the multilingual model on their own hardware. Just a few weeks after the Jev model was announced, Microsoft Decision, OpenAI Decisions API, and Fastino Glide emerged. The parameter files for the Cloudflare Clef and Perplexity Decider models were released publicly. As a result, users with suitable hardware can run them on their own systems. Open-source alternatives include Amazon Strands Decider 2B, Kev, Von, Jeff, and jevlike. The list also includes SemIf, OpenJev, NanoJev, and Ollaya. TypeSafe and Mukkunnoth have called their models System One and System 1. These names are believed to have been taken from Daniel Kahneman’s concept of fast thinking. Most rivals add layers for input, output, and decision-making mechanisms to Alibaba’s open-source Qwen model. Laya is not an LLM. It is based on the BERT family of models, which do not generate text and only encode information. Although the projects’ rapid proliferation raises suspicions, it does not prove that the research was copied.
Jev works as an API. An API is an interface used for software to connect to this service. The system receives a customer’s online purchase information and submitted questions. For example, it can determine whether the buyer wants to return a product and which product the buyer is referring to. The response is provided as a probability score for yes-or-no or multiple-choice options. The developer then uses that number to set conditions in the software. Because the model does not generate text, it does not need previous conversations or extensive context. It processes questions simultaneously and does so at a lower cost than conventional large language models (LLMs). TypeSafe also published a brief technical explanation of the System One architecture. Just a few days later, Mukkunnoth wrote on his blog that he had developed the Jev architecture 1 year earlier. He published a scientific paper about SalesRLAgent, a model that calculates probabilities, on March 30, 2025. He also openly shared the model’s ready-to-use parameter files and dataset. A 2nd study released in September was closer to how the Jev model works. According to Mukkunnoth, SalesRLAgent was overlooked because it only solved a specific problem in sales. TypeSafe turned the same approach into a decision engine that can work across different fields. The finished product includes the company’s servers, API access, documentation, developer tools, and payment plans. Mukkunnoth later released the Laya model as open source. Users can run the multilingual model on their own hardware. Just a few weeks after the Jev model was announced, Microsoft Decision, OpenAI Decisions API, and Fastino Glide emerged. The parameter files for the Cloudflare Clef and Perplexity Decider models were released publicly. As a result, users with suitable hardware can run them on their own systems. Open-source alternatives include Amazon Strands Decider 2B, Kev, Von, Jeff, and jevlike. The list also includes SemIf, OpenJev, NanoJev, and Ollaya. TypeSafe and Mukkunnoth have called their models System One and System 1. These names are believed to have been taken from Daniel Kahneman’s concept of fast thinking. Most rivals add layers for input, output, and decision-making mechanisms to Alibaba’s open-source Qwen model. Laya is not an LLM. It is based on the BERT family of models, which do not generate text and only encode information. Although the projects’ rapid proliferation raises suspicions, it does not prove that the research was copied.