In the wave of artificial intelligence, innovations in chips and software have always been key drivers of industry development. Now, emerging AI company Recogni has introduced a revolutionary computing method that not only makes AI chips smaller and faster but also significantly reduces operating costs.
Recogni's new technology is named "Pareto," which adopts an innovative logarithmic approach, enabling it to outperform existing technology in performance metrics when running large AI models. This breakthrough has received support from renowned companies and venture capital firms such as BMW, Bosch, and Mayfield.
Traditional AI models, such as OpenAI's GPT-4 and Google's Gemini, require extensive power-consuming computations when handling simple chat prompts. Recogni's Pareto system significantly reduces energy consumption by converting multiplication operations into addition operations, while maintaining the accuracy of the model.
Image source note: The image is generated by AI, authorized service provider Midjourney
Recogni has tested the Pareto system on AI models developed by companies like Meta and Stability AI, achieving positive results. Currently, the company's first chip is designed and manufactured using Taiwan Semiconductor Manufacturing Company's seven-nanometer process, and is collaborating with an undisclosed partner to announce a broader rollout plan in the coming months.
Gilles Backhus, co-founder and AI Vice President of Recogni, stated that the company is in talks with some companies that offer data center hardware leasing services, which will be an important direction for the deployment of the Pareto technology.