The secret ingredients powering AI chatbots

Bill Ready, CEO of Pinterest, emphasized during their earnings call that using open-source AI models saves the company a significant amount of money, less than 8% compared to proprietary systems. This cost-saving is attributed to Chinese models, such as Alibaba's Qwen, which is fine-tuned using Pinterest's data and supports shopping assistants and chatbots.

Several American companies, including DoorDash, have also adopted Chinese AI models for cost savings and to develop consumer-facing tools like chatbots for trip planning, product recommendations, and processing refunds. OpenRouter offers insight into the popularity of different AI models among developers, but the use of Chinese models by large American companies remains largely undisclosed.

This trend has heightened tensions between US businesses and the government over concerns about safety and national security related to Chinese AI. Surveys indicate that Americans trust American AI models more than Chinese ones, with only 9% trusting Chinese models and 52% trusting American models. Users may be unaware if their conversations with a brand's chatbot are with AI developed in China or the US, a distinction that matters to Americans.

While executives face pressure to demonstrate prudent spending on technology, the opaque nature of AI stacks makes it challenging to determine the exact models used for chatbots. Despite concerns about potential biases and geopolitical implications, companies remain hesitant to disclose the specific models powering their chatbots, even under pressure from lawmakers and government officials. Transparency in AI model usage, however, is crucial for users to make informed decisions about their interactions with AI systems.