GPT-4 Sets a New Benchmark for Large Language Models
In Brief
OpenAI has officially unveiled its latest AI model, GPT-4.
GPT-4 outperforms all existing large language models , boasting an exceptional performance score of 85.5% in English language proficiency.
On March 14, OpenAI announced the arrival of GPT-4, a significantly upgraded successor to its GPT-3.5. This model has fueled the popular ChatGPT chatbot since November 2022. Alongside the announcement, OpenAI also presented a demonstration showcasing the capabilities of its new Generative Pre-trained Transformer.
Mira Murati, OpenAI's Chief Technology Officer, commented that GPT-4 demonstrates human-like capabilities across numerous professional and educational standards, emphasizing that after extensive alignment and adversarial testing, it stands out as the best model we have created concerning factual accuracy, steerability, and safety. shared The revamped AI model showcases
a remarkable improvement over GPT-3.5 on various analytical standards. For example, the ChatGPT utilizing the GPT-3.5 framework achieved a score of 70.1% in English, while GPT-4 excelled with a score of 85.5% in achieved multitasking language comprehension. Moreover, the new AI displayed impressive results in additional languages. For instance, it scored 84.1% on an Italian test and achieved a 73.2% in the Bengali language assessment. Even in low-resource languages like Swahili, Latvian, and Welsh, the performance exceeded that of GPT-3.5 English results. in English.
Interestingly, the popular language-learning tool Duolingo
has incorporated GPT-4 into its latest subscription service, \"Duoling Max.\" This AI is set to deliver highly customized language lessons and affordable, accessible English proficiency assessments. implemented As a quick note, users can access GPT-4 through the premium ChatGPT Plus subscription. Meanwhile, developers interested in utilizing OpenAI's API to build their applications and services are encouraged to register for the
GPT-4 Demonstrates Its Superiority Over GPT-3.5 on Multiple Learning Assessments waitlist .
Read more about GPT-4:
Disclaimer
In line with the Trust Project guidelines Valeria is a journalist for Metaverse Post, focusing on fundraising, AI, the metaverse, digital fashion, NFTs, and all things related to web3.
