Zaima Mashiat Nabi
Recently the artificial intelligence research lab OpenAI announced the release of a new model called GPT-4 which claims to be much more potent than the former version. This version is said to be much more innovative and less prone to make any mistakes regarding facts.
According to the co-founder of OpenAI, GPT-4 takes images as input and answers all the questions regarding the given image alongside texts. This version has the potential of recollecting and answering up to 25,000 words at once. Several of the complaints users had about the previous iteration of OpenAI’s system, according to OpenAI, have been addressed or improved in GPT-4.
However, GPT-4 will still hallucinate facts for which the owners have alerted us that great care must be taken while working with very important and risky content. Furthermore, this new system would not work as well as the original ChatGPT in terms of cliches.
GPT-4 might also make a few arithmetic errors which a calculator would never do. As we know it takes images as input for analysis, but GPT-4 is still not strong enough to examine video and audio inputs. Additionally, GPT-4 is not quite autonomous and only works according to human commands.
GPT-4 is said to be much more multilingual as it handles around 26 languages other than English. This version is programmed to work with different personalities due to steerability, a new and attractive update to the system.
To have full access to the GPT-4 version we need to have a paid subscription to OpenAI which has a monthly cost of $20 only. The performance of the system on various assessments and standards, including the LSAT, SAT Math, and SAT Evidence-Based Reading & Writing exams, is claimed to demonstrate GPT-4’s advances.
This system has been safe tested for 6 months for which now it will omit 82% of the disallowed content and will emit generate 40% much more accurate facts. All these made the responses of GPT-4 to be much more humanlike and precise.
