Four leading companies in the field of artificial intelligence — OpenAI, Google, Anthropic, and Microsoft — have encountered serious difficulties in creating new language models. According to Bloomberg, the problem lies in the inability to significantly improve results compared to previous versions.
OpenAI completed training its Orion model back in September, but its performance turned out to be disappointing. According to sources, “the leap from GPT-4 to Orion is much smaller than from GPT-3.5 to GPT-4.” The model cannot effectively respond to new queries, which is why its launch has been postponed to 2025.
A similar situation is unfolding at Google and Anthropic. Google is working on its Gemini model, which so far does not meet expectations. Anthropic has also delayed the release of Claude 3.5 Opus due to weak test results.
Recently, Microsoft has also faced difficulties. In October, the company introduced an update for Copilot which, according to Windows Central, received negative user feedback. Users noted a decline in response quality and even requested a return to the previous version. Insiders report that “there is a gap between Microsoft’s ambitious vision and the actual user experience.”
The main problems for all these companies are related to a lack of high-quality training data and rising development costs. Developers admit that the first models were easily trained on open sources, but now more complex and unique data are needed. In addition, investments in new models are becoming riskier due to the lack of significant progress.
