Ecosia, a Berlin-based search engine used by government agencies, has replaced French AI firm Mistral with open-source models, including Chinese ones, citing quality issues and sovereignty concerns. The switch has reportedly cut Ecosia's costs in half while improving performance.
TL;DR
- Ecosia replaces Mistral with open-source models, including Chinese ones, due to quality and cost concerns.
- The switch has halved Ecosia's costs while improving performance, according to CEO Christian Kroll.
- Ecosia's move raises questions about Europe's AI sovereignty and the viability of open-source models.
What happened
Ecosia, a Berlin-based search engine used by institutions like Germany's federal environment ministry and the U.K.'s National Health Service, has replaced French AI firm Mistral with open-source models, including Chinese ones. The decision comes after Ecosia experienced recurring technical problems with Mistral, including overloaded servers, according to CEO Christian Kroll.
Ecosia had initially replaced OpenAI with Mistral in May, but Kroll expressed disappointment with Mistral's model quality, stating that they are 'a year behind' the competition. Additionally, Ecosia has concerns about Mistral's alignment with its environmental goals and European sovereignty, given Mistral's international investments.
Mistral, however, has recently released a new model, Large 4, and chief scientist Guillaume Lample encouraged Ecosia to test it, offering early access. Mistral has raised billions of euros from international investors, including Samsung and Nvidia.
Why it matters
Ecosia's switch to open-source models raises questions about Europe's AI sovereignty and the viability of open-source models in the face of competition from the U.S. and China. The move also highlights the challenges of building a competitive, sovereign AI model on European soil.
For AI/ML developers and startups, this shift underscores the importance of model quality and performance, as well as the potential cost savings and customization offered by open-source models. It also raises questions about the geopolitical implications of using Chinese AI models, given concerns about censorship and bias.
For tech investors, Ecosia's move to open-source models could signal a shift in the AI landscape, with open-source models gaining traction as a cost-effective and customizable alternative to proprietary models. However, the geopolitical risks associated with Chinese models may also present challenges for investors.
Key facts
- Ecosia has replaced Mistral with open-source models, including Chinese ones, citing quality issues and sovereignty concerns.
- The switch has reportedly cut Ecosia's costs in half while improving performance, according to CEO Christian Kroll.
- Ecosia is partnering with the AI platform Melious, which hosts a range of open-source and open-weight models, including Chinese models Qwen, GLM, and Kimi.
- Mistral had raised billions of euros from international investors, including Samsung and Nvidia.
- Mistral's chief scientist Guillaume Lample encouraged Ecosia to test their new model, Large 4, offering early access.
- Ecosia had initially replaced OpenAI with Mistral in May, but experienced recurring technical problems.
- Ecosia is used by institutions like Germany's federal environment ministry and the U.K.'s National Health Service.
- Some experts argue that subtle bias in Chinese AI models is more difficult to address than overt censorship.
Context
The AI landscape is increasingly dominated by a few major players, including OpenAI, Mistral, and Chinese companies like Baidu and Alibaba. The competition between these players is intense, with each seeking to gain an edge in model quality, performance, and cost.
Open-source models have gained traction in recent years as a cost-effective and customizable alternative to proprietary models. However, concerns about the quality, performance, and geopolitical implications of open-source models, particularly those from China, remain.
The European Union has been seeking to reorient its AI strategy, with some leaders and advisors advocating for a focus on open-source models as a way to build on the continent's strengths and avoid the high costs of training frontier systems.
