Inflection AI has unveiled its second-generation large language model (LLM), Inflection-2, positioning itself as a formidable competitor to industry giants Google and Meta. Trained on an impressive 5,000 NVIDIA H100 GPUs with approximately 10^25 FLOPs, Inflection-2 boasts performance levels comparable to Google’s PaLM 2 and even outperforms it in benchmarks such as MMLU, TriviaQA, HellaSwag, and GSM8k.
Notably, Inflection-2 stands as the world’s second most capable LLM, trailing only behind OpenAI’s GPT-4. Despite its size, this powerhouse model reduces costs and increases serving speed compared to its predecessor, Inflection-1, thanks to its transition from A100 to H100 GPUs, showcasing remarkable training efficiency.
Designed with a focus on serving efficiency, Inflection-2 is slated to power the Pi chatbot, marking a significant milestone in the quest to create a personal AI accessible to everyone. The company has worked with NVIDIA, Microsoft, and CoreWeave in building the AI cluster.
Inflection AI underscores its commitment to safety, security, and trustworthiness in very large model training. The model aligns with global ethical standards and has voluntarily committed to the White House’s ethical AI guidelines set in July 2023.
Benchmark results showcase Inflection-2’s superiority over its predecessor and external models, such as LLaMA-2, Grok-1, PaLM-2, Claude-2, and GPT-4. Particularly noteworthy is its exceptional performance in chain-of-thought reasoning, outshining even Claude 2 in evaluations on MMLU (5-shot).
CEO Mustafa Suleyman emphasizes the importance of AI safety and ethical standards, urging empathy and forgiveness in light of recent challenges faced by rivals like OpenAI. Despite acknowledging the competitive nature of the industry, Suleyman hopes for understanding and collaboration, as Inflection-2 prepares to usher in a new era of innovation in the rapidly evolving Silicon Valley landscape.
What is Inflection AI:
Inflection AI, which was founded in 2022 is a pioneering technology company structured as a public benefit corporation. Renowned for its collaboration with Nvidia in developing hardware for generative artificial intelligence, the company made headlines in June 2023 by securing a substantial $1.3 billion USD in funding. Their flagship product, Pi, is a chatbot designed to serve as a personal intelligence assistant. Boasting a unique user experience focus, Pi aims to provide emotional support through interactive text-based dialogues with users, incorporating elements of kindness, diplomacy on sensitive topics, and humor. This innovative venture draws comparisons with OpenAI’s ChatGPT, showcasing Inflection AI’s advancements in the frontier of artificial intelligence applications.
Related:
- OpenAI Drama Unfolds Amidst AGI Progress and Safety Concerns
- OpenAI has made ChatGPT Voice Available to Everyone Amidst Internal Tensions
- OpenAI resolves leadership crisis: Sam Altman reinstated as CEO amidst governance overhaul
