Company Valuation Reaches 7.5 Billion USD Following Funding Round

The artificial intelligence model 'Jev,' developed by TypesafeAI, has achieved a company valuation of 7.5 billion USD just weeks after its launch.

According to TechCrunch, TypesafeAI secured 870 million USD in a funding round led by Andreessen Horowitz (a16z), with participation from Sequoia and existing investor DCVC.

Jev gained rapid attention following its release on September 15, with TypesafeAI claiming that one-third of Fortune 500 companies are already utilizing the model.

Founded in 2024, TypesafeAI was co-founded by CEO Diogo Almeida, who previously worked at OpenAI, COO Sasha Cheng, a former research engineer at Meta, and CTO Eric Gafni, an engineer and entrepreneur.

While Jev is based on transformer architecture, it is not a large language model (LLM). Instead of generating text, it produces probabilistic outputs, which the company refers to as 'calibrated decisions.'

Almeida noted his experience at OpenAI developing ChatGPT and researching human feedback-based reinforcement learning (RLHF), which highlighted the limitations of LLM capabilities in actual automation.

Jev addresses this problem by providing values that software can directly process, such as probabilities for yes/no questions, selections from lists, or scores, all with defined certainty levels. Users can specify the output range in advance, allowing other programs to use the results without additional interpretation.

TypesafeAI claims this approach increases model speed, reduces costs, and prevents hallucinations that can occur with generative models.

The model is named after economist William Stanley Jevons, known for his 'Jevons Paradox,' which suggests that lower costs of a resource can lead to increased usage. Almeida expresses hope that as the cost of intelligence decreases, intelligent software will become more widespread.

He emphasizes the importance of creating all necessary data in-house, noting that half of the company is dedicated to researching statistically sound synthetic data.

Reported by ZDNet Korea.