Inflection AI has been creating a buzz in the world of large language models (LLMs) with the launch of Inflection-2.5, a model that competes with leading LLMs like OpenAI’s GPT-4 and Google’s Gemini. The company’s rapid growth has been supported by a significant $1.3 billion funding round from industry giants such as Microsoft and NVIDIA, as well as investors like Reid Hoffman, Bill Gates, and Eric Schmidt. This funding brings the total raised by Inflection AI to $1.525 billion.
In partnership with CoreWeave and NVIDIA, Inflection AI is constructing the largest AI cluster globally, consisting of 22,000 NVIDIA H100 Tensor Core GPUs. This massive computing power will aid in training and deploying a new generation of large-scale AI models, allowing Inflection AI to push the boundaries of personal AI.
The company’s work has already produced impressive results, with the Inflection AI cluster – currently with over 3,500 NVIDIA H100 Tensor Core GPUs – achieving outstanding performance on the open-source MLPerf benchmark. In collaboration with CoreWeave and NVIDIA, the cluster completed the reference training task for large language models in just 11 minutes, establishing itself as the fastest cluster on this benchmark.
Following the introduction of Inflection-1, Inflection AI’s proprietary large language model, which has been lauded as the top model in its compute class, the company’s commitment to transparency and reproducibility is evident in a technical memo detailing the evaluation and performance of Inflection-1. The memo reveals that Inflection-1 outperforms models in the same compute class, defined as models trained using at most the FLOPs of PaLM-540B.
Inflection-2.5 is now available to Pi users, Inflection AI’s personal AI assistant, across various platforms, including the web, iOS, Android, and a new desktop app. This integration marks a significant step in Inflection AI’s mission to create a personal AI for everyone, combining raw capability with an empathetic personality and safety standards.
Inflection-2.5 represents a leap in performance, particularly in coding and mathematics. The model excels in industry benchmarks, showcasing over 94% of GPT-4’s average performance across various tasks, with a focus on STEM areas. Inflection-2.5 outperforms Inflection-1 on coding benchmarks like MBPP+ and HumanEval+, demonstrating its dominance in the coding domain.
In STEM examinations, Inflection-2.5 shines on the Hungarian Math exam and Physics GRE, showcasing its mathematical aptitude and problem-solving skills. With its powerful capabilities, Inflection-2.5 enhances user experience by offering high-quality, up-to-date information and guidance across a wide range of topics.
The integration of Inflection-2.5 into Pi has led to increased user adoption and engagement, with one million daily and six million monthly active users exchanging over four billion messages with Pi. Inflection AI’s commitment to transparency is evident in the detailed technical results and benchmark performance of Inflection-2.5.
As Inflection AI continues to innovate and push the boundaries of LLMs, the company’s visionary approach and dedication to creating high-quality, safe AI experiences position it as a trailblazer in the AI landscape.
Source link
