AI start-up Inflection AI recently presented its latest generative AI LLM Inflection-2. This model could easily compete with almost all other known generative AI models, according to the AI developer.
The now presented generative AI model Inflection-2 is a big improvement over the first Inflection-1 model, according to Inflection AI. The latter model’s capabilities include writing texts, rewriting, summarizing documents, generating code and performing related tasks.
Inflection-2 now offers more capabilities. This is partly due to a significantly improved knowledge database, which makes it capable of answering user queries more accurately. Also, the updated generative AI model is said to reason better. This would, for example, generate better code or make texts more detailed.
Inflection AI immediately subjected its latest generative AI model to a benchmark with competitors. According to the AI specialist, these tests show that it often outperforms competitors’ existing generative AI models.
The benchmark showed that Inflection-2’s performance would exceed Google’s PaLM 2, Anthropic’s Claude 2, Meta’s Llama 2 and the “rebellious” Grok AI model. Only against OpenAI’s GPT-4 model does Inflection-2 still lose out.
Trained with Nvidia H100 GPUs
To train the Inflection-2 model, the AI specialist used as many as 5,000 Nvidia H100 GPUs. In addition, these specific GPUs are used for inference. The AI developer has a total of 22,000 Nvidia H100 GPUs at its disposal for generative AI LLMs.