How GPT4 is different from ChatGPT or GPT3.5 – All you need to know

GPT 4 vs GPT 3.5 or ChatGPT: Micrsoft-backed OpenAI created a stir when by releasing the next version of its superhit Generative AI model – GPT4. ChatGPT was built on GPT3.5, where GPT stands for Generative Pre-trained Transformer. As the name (Chat) suggests it was a deep learning technique that employs artificial neural networks to produce human-like writing. Recently, OpenAI announced its new large multimodal model ‘GPT4’ which accepts image and text inputs. The company also released a $20 per month ChatGPT Pro based on GPT4. 

How GPT4 is different from ChatGPT or GPT3.5:

1. It is more precise than GPT3.5 (ChatGPT), but still has flaws and still ‘hallucinates’, though to a lower extent. According to Jaspreet Bindra, Founder The Tech Whisperer Limited, UK, larger training data sets made this possible. 

“ChatGPT was short on facts since it was designed to be plausible rather than truthful. GPT4 has a similar design principle but has been tweaked such that it has more true facts in it, which ChatGPT is lacking. OpenAI also launched ChatGPT Pro which at $20 a month gives a better, faster experience and this bot is built on GPT4. ChatGPT might continue to be free, but built on GPT3.5 with a more degraded experience,” Bindra said.

2. GPT4 can describe images, say the Mona Lisa painting, in great detail. The GPT4 demo by OpenAI showcased its astonishing ability to build a website on the fly, from a drawing and text written on paper. 

ALSO READ | Bard Vs ChatGPT: Difference between ChatGPT and Bard, who has got most features

It also has the ability to describe a painting in the level of detail that you ask it to, including the artist and what the painting depicts. This could be a game-changer. 

“GPT4 is rumoured to be totally multimodal, so it could potentially do this for images, video and music. OpenAI is still wary of the potential for misuse, so it has not released it for the general public yet,” Bindra said.

3. GPT4 has greater accuracy and reasoning capabilities. It has passed impressive exams including SAT (a test widely used for college admissions in the United States) and the Bar exam; it passed a simulated Bar exam with a score around the top 10 per cent of test takers. On the other hand, GPT3.5’s score was in the bottom 10 per cent. This is the other astonishing bit. GPT4 passed the SATs with 1700 out of 2400 accounts. 

4. It still does not do recent events well, since the training on the data sets stopped around late 2021. So there is no noticeable difference here. It also does not learn from experience. However, on the data until 2021, GPT4 does better in terms of reasoning, accuracy and the quality of results

5. GPT4 has very impressive business use cases. Khan Academy is using it as a tutor to make children learn. A company called BeMyEyes built a GPT4-powered ‘Virtual Volunteer’ for blind and low-vision users. Morgan Stanley has started using GPT4 for wealth management. Harvey, a UK startup, has signed a deal with PWC and Bain for internal knowledge management. The forthcoming ability to create a website out of scribbled instructions and drawings will be another powerful use case. 

6. It is multilingual. “Using Microsoft Azure’s capabilities to translate and converse in multiple languages. GPT4 was tested on its reasoning abilities across 26 languages (including Bengali!) and it exhibited superior reasoning than GPT3.5 /ChatGPT in 24 of them,” Bindra.