Inflection AI, the startup behind the chatbot Pi, has released a new AI model called Inflection-2. The company claims that it outperforms Google’s PaLM Large 2 model and Meta’s LLaMA 2 model on various benchmarks. However, it still trails behind OpenAI’s flagship model, GPT-4. The new model will be integrated into Pi after some additional work. Inflection CEO Mustafa Suleyman believes that they are just at the beginning of the scaling curve for AI models. The release of Inflection-2 comes amidst the turmoil at OpenAI, with CEO Sam Altman being temporarily fired and then reinstated. Inflection insists that the release was not influenced by the events at OpenAI. The startup used 5,000 Nvidia GPUs to train the new model, which is faster and cheaper to train than its predecessor. Inflection-2 performed well on various benchmarks, positioning it as one of the top models in its size category. Inflection plans to focus on training an even larger model in the future. Suleyman urges empathy and forgiveness towards those involved in the OpenAI board battle. He also acknowledges the competitive nature of the AI industry and the opportunity it presents for his business.
