Inflection, an AI startup aiming to create “personal AI for everyone,” has announced a new large-scale language model called Inflection-2 that outperforms Google's PaLM 2.
Inflection-2 was trained on over 5,000 NVIDIA GPUs and reached 1,025 quintillion floating point operations (FLOPs), the same level as PaLM 2 Large. However, early benchmarks show that Inflection-2 outperforms Google's model on tests of reasoning ability, factual knowledge, and stylistic proficiency.
Across a variety of common academic AI benchmarks, Inflection-2 achieved higher scores than PaLM 2 on most benchmarks. This includes outperforming the search giant's flagship tests on the TriviaQA, HellaSwag, and Grade School Math (GSM8k) benchmarks, as well as the diverse Multi-Task Middle School Language Understanding (MMLU) test.
The startup's new model will soon enhance the personal assistant app Pi, enabling more natural conversations and useful features.
Even though Inflection-2 is much larger than previous versions, the move from NVIDIA A100 to H100 GPUs for inference (combined with optimization efforts) results in faster processing and lower costs. He said it would be reduced.
An Inflection spokesperson said the latest model brings the company “closer to a major milestone” in achieving its mission of bringing AI assistants to everyone. They added that the team is “already looking forward” to training even larger models on his 22,000 GPU supercluster.
Safety is said to be a top priority for researchers, and Inflexion is one of the first signatories of the White House's July 2023 Voluntary AI Commitment. The company said its safety team continues to work to rigorously evaluate the model and ensure it relies on best practices for adjustments.
With impressive benchmarks and further expansion plans, Inflection's latest effort poses a serious challenge to tech giants like Google and Microsoft, which have traditionally dominated the large-scale language modeling space. The race to realize the next generation of AI continues.
(Photo by Johann Walter Bantz on Unsplash)
See also: Anthropic upsizes Claude 2.1 to 200,000 tokens, nearly doubles GPT-4
Want to learn more about AI and big data from industry leaders? Check out the AI & Big Data Expos in Amsterdam, California, and London. This comprehensive event coincides with Digital Transformation Week.
Learn about other upcoming enterprise technology events and webinars from TechForge here.
