Deep Cogito raises $43M to develop self-improving AI models

Deep Cogito, an artificial intelligence startup founded in 2024 by former Google LLC employees Drishan Arora and Dhruv Malrana, has secured $43 million in funding to develop self-improving AI models. The Series A round was led by TQ Ventures, with participation from Benchmark, Nexus Venture Partners, Atreides Management, South Park Commons and Zscaler Inc. This brings Deep Cogito's total outside funding to over $56 million.

The company's research focuses on creating AI models that can autonomously improve themselves. Their efforts have resulted in the Cogito family of open-source large language models, with the latest version, Cogito v2.1 671B, launched in November. Deep Cogito claims this model outperforms other U.S. open models available at the time while using less computational power.

To enhance model quality, Deep Cogito employs a technique called IDA, which temporarily increases computational resources during an LLM's response generation. The company also uses reinforcement learning, a method where an LLM receives feedback on task performance from a second neural network. Deep Cogito's unique process supervision approach evaluates the steps an LLM takes to reach its answer, reducing unnecessary or incorrect steps and saving hardware resources.

The new funding will be utilized to expand Deep Cogito's enterprise business, grow its research team, and release additional open-source models. The company's technology is monetized through a platform that enables businesses to create custom AI models using internal data. Zscaler, an investor in the round, is one of the platform's users.

Deep Cogito joins other startups, such as Recursive Superintelligence Inc., which raised $650 million in May from a consortium that included Nvidia Corp., and Ineffable Intelligence Ltd., which closed a $1.1 billion round in April at a $5.1 billion valuation.