Open AI’s ChatGPT 3 has become too popular amongst various industries in the last few days. Whether we talk about the tech or medical sectors, the AI Chatbot is doing impressive work. People are getting too many benefits by using the GPT-3. And now, Open AI has improved it and launched a new version of the AI Bot called GPT-4. Undoubtedly, this is far better than the older GPT-3, but still, it has some fundamental flaws that researchers feared at the launch of GPT-3. Here we will tell you five reasons why GPT-4 might be a game changer but will not be perfect and compete with the human race.
Improvement in Accuracy.
Oren Etzioni – an A.I. researcher and professor, has tried to compare the GPT-3.5 and GPT-4 by asking a straightforward question – “What is the relationship between Oren Etzioni and Eli Etzioni?” The New bot has responded correctly almost.
The earlier version of the chatbot responded incorrectly but got much more accurate answers from the new bot. But it still had some mistakes, such as all the information provided by it was correct, except it didn’t have data about Dr. Etzioni having stepped down from the position of Chief Executive of Allen institute.
GPT-4 can describe images with impressive detail.
Greg Brockman, president and co-founder of OpenAI, has demonstrated how GPT-4 will also respond to the images. It has shown how the AI could precisely describe an image of the Hubble Space Telescope.
Further, if you show it inside your fridge, It can advise you what you can make from the items in your refrigerator. This service has not been made publically yet, but a company named Be My Eyes is using the bot to build more services to show details from these types of images from the internet or real-time snaps.
GPT-4 with some severe expertise
Anil Gehi, an associate professor of medicine and a cardiologist at the University of North Carolina at Chapel Hill, tried this bot. He filled in the medical history of the patient and the current situation of the patient. And asked the bot to provide the details of how they should treat the patient and were shocked to know when the chatbot provided the exact details they would use to treat the patient.
But it is not apparent that the bot will always provide some accurate information. It still needs experts for review once.
It can reason — up to a point.
Dr. Etzioni has given a puzzle to the bot, and it seems like the earlier version of GPT was better than the new one as it had reasoning skills. GPT-4 has responded to the puzzle but didn’t consider the height-width parameters, which can eventually prevent the car and tank from going through the door.
Not good at discussing the future
Through the research, it has been found that GPT-4 is good with historical information, but if we talk about the future, then it cannot formulate a new plan. It can’t make future decisions even if it can make images as per the instructions, but it can’t do anything new and creative.
Further, it has been noted that the new chatbot hallucinates as it does not know the difference between true and false. So when asked for the website’s address for the latest cancer research, it provides some unavailable addresses.
So that are some of the flaws encountered during the initial research. No doubt the GPT-4 is 500 times more potent than the GPT-3, but still, it needs a lot of improvements. What do you think about how this new version of AI chatbot will revolutionize different sectors?