Gemini Ultra's design allows it to process several types of data, from text to audio to code to images and video, which Google claims gives it an advantage over other models. However, GPT-4 did outperform Gemini Ultra in common sense reasoning abilities for everyday tasks. Early reactions to the less advanced Gemini Pro have been positive, but it has had issues with accuracy and hallucinations. It remains to be seen how the trio of Gemini models will fare against OpenAI's GPT-4.
Key takeaways:
- Google unveiled Gemini, a rival to OpenAI's GPT-4, which outperforms GPT-4 in math, coding, and subject knowledge at its most advanced tier.
- The most advanced model, Gemini Ultra, is not available to the public yet, but it outperforms GPT-4 in several areas and is the first model to outperform human experts on the Massive Multitask Language Understanding (MMLU) test.
- Gemini Ultra is natively multimodal, designed to process several types of data, from text to audio to code to images and video.
- Early reactions to the less advanced Gemini Pro, accessible through Google's chatbot Bard, have been positive, but there have been issues with accuracy and hallucinations.