Big Tech companies and AI startups spend hundreds of millions of dollars training their language models – and releasing them in whole or in part. The weights are usually published using an open weights approach. This makes it possible to operate AI models on your own devices, autonomously, permanently free of charge and data-efficient.
Meta has disclosed the weights of its language model Llama, and the French AI startup Mistral and the Chinese IT group Alibaba do the same (with Qwen). Google (with Gemma), Microsoft (Phi) and OpenAI (gpt-oss) have released small versions of their language models.
Dirk Riehle has a professorship for open source software at the Friedrich-Alexander University Erlangen-Nuremberg. There he deals, among other things, with classic open source business models. We asked him to explain the strategies behind opening up AI models.
That was the excerpt from our heise-Plus article “Why Big Tech companies are opening up their AI models – an interview”. With a heise Plus subscription you can read the entire article.
