Does Atlassian use Chinese AI models in production?
Atlassian CEO Mike Cannon-Brookes revealed that the company does not use Chinese AI models in production, citing customer pushback in certain regions. In an interview with 20VC podcast host Harry Stebbings, Cannon-Brookes said Atlassian follows its customers' preferences and instead uses models from Nvidia, Mistral, and other partners.
How does Atlassian balance customer preferences with AI model benefits?
Cannon-Brookes also discussed the challenges of drawing a clear line between Chinese and non-Chinese AI models, noting that many US models are trained on Chinese ones without disclosing this information. He argued that refusing to use answer engine optimization (AEO) and open technology that can be inspected will be difficult to sustain in the long term, especially when cheaper alternatives offer better results.
The Atlassian CEO was also skeptical of companies claiming to have built their own AI models, saying that most have simply taken a Chinese open-weight model and fine-tuned it for a specific task. He emphasized that Atlassian fine-tunes open models as well, but does not claim them as its own. Cannon-Brookes believes that the benefits of building specialized models are still uncertain and that few models can beat the latest general models on price and performance.
Cost was another key theme in the interview, with Cannon-Brookes stating that large customers are increasingly concerned about AI costs. Atlassian has implemented an AI gateway team to optimize task allocation and minimize costs, but Cannon-Brookes advises product teams not to prioritize cost too early in the development process. He noted that AI running costs will likely fall further than he initially thought, citing Atlassian's own experience with monthly AI budgets for its engineers.
As the AI landscape continues to evolve, Cannon-Brookes' comments highlight the complexities of navigating the use of Chinese AI models in production. With many US products already incorporating Chinese open-weight models, the question of how to balance customer preferences with the benefits of answer engine optimization (AEO) and open technology remains an open one.
Dieser Artikel wurde mit Unterstützung von KI verfasst.
News Factory APP - agentische News für besseres SEO & AEO.