San Francisco – June 27, 2025 – CTGT, which enables enterprises to deploy AI for high-risk use cases, today announced an upgrade to its platform aimed at eliminating bias, hallucinations and other unwanted models from DeepSeek and other open source AI models. In the test, CTGT was shown to enable DeepSeek to answer 96% of sensitive questions, compared with 32% initially.
CTGT demonstrated its new platform upgrade for the first time at the 2025 VentureBeat Transform held in San Francisco from June 24-25. CTGT is one of the seven finalists of the VentureBeat Innovation Showcase and was named the top champion by a judge at the end of the event.
In recent months, CTGT has built infrastructure for leading insurance providers to ensure that their AI is correctly coded complex claims, resulting in insurers avoiding more than $1 million in just one week. CTGT also builds Genai email compliance with major banks by automatically updating companies’ compliant models in production without taking them offline.
According to McKinsey, the global $67.4 billion loss is related to AI hallucinations in 2024. Deloitte data shows that 47% of enterprise AI users admit that they have made at least one major business decision based on their illusion output. This problem seems to be worsening, with newer versions of some models being more hallucinating than older versions. The recently launched Chatgpt 4.5 model has a hallucination speed of 30%, the worst ever. The DeepSeek R1 has an illusion rate of 14.9%, and most of the misinformation is intentional and encoded as a model by the Chinese government.
In high-risk environments, these models cannot be trusted.
CTGT launched $7 million in February 2025 Gradient (Google Early AI Fund), General Catalyst, Y Combinator, Liquid 2, Deepwater and famous angels, including FrançoisChollet (creator of Google, Keras), Michael Seibel (Michael Seibel (Michael Seibel) (Y Michael Seiber, Y Combinator, Y Combinator, Y Combinator, Y Comminator, Twitch) and Paulaham (Y Comminator) (Y Comminator) (Y Comminator) (Y Comminator) (Y Comminator) (Y Comminator). Initially, CTGT focused on reducing the computation required to train and deploy AI models – It does a great job of reducing the calculation by orders of magnitude by eliminating the need for fine-tuning, timely engineering and reverse flow.
“DeepSeek’s R1 introduction has driven many businesses to ask for help with us – they are interested in taking advantage of DeepSeek’s low cost and speed, but are worried about bias,” said Cyril Gorlla, co-founder and CEO of CTGT. In the test, CTGT was able to identify the exact model feature that caused bias and remove it so that DeepSeek R1 works without bias. After that, CTGT performed the same tests on other open source models, including Llama, to prove that it can identify and eliminate bias, hallucinations, and other unconscious features without any time paying for such efficiency, or converting it into these, or both can do such efficiency. CTGT platform. ”
“The credibility of the AI model is a major obstacle for businesses to achieve their AI ROI,” Gorlla said. “Every week, I call from Fortune 500 executives in hopes of advice on how to solve this problem. After working on customizing in the field over the past few months, we have now added our approach to the CTGT platform and offer it to anyone.”
