Identifying the best models in the field of artificial intelligence is difficult due to the versatile and complex nature of this technology. The data sets on which the models are trained and the structure of this data directly affect the results. That’s why the most accurate method when ranking artificial intelligence models is to look at the accuracy rate of the outputs they provide. In this context, the AI hallucination table published by Vectara ranks various AI chatbots according to their ability to ‘not hallucinate’.
Artificial Intelligence’s Hallucination Problem and Sorting
AI models are prone to ‘hallucination’, a known phenomenon. This means that models sometimes fill in the gaps by producing information that is not factual. This process is so finely tuned that an unattentive user can easily be misled.
According to Vectara’s evaluation, the model with the lowest hallucination rate and highest accuracy was determined as GPT-4. This model is followed by GPT-3.5 developed by OpenAI. At work best artificial intelligence models:
- GPT 3.5
- Llama 2 70B
- Llama 2 7B
- Llama 2 13B
- Anthropic Calude 2
- Mistral 7B
- Google Plam-Chat
As artificial intelligence technology develops rapidly, the accuracy rates and reliability of these models also gain importance. Vectara’s evaluation shapes the future direction of artificial intelligence and users’ trust in this technology.