The generative AI chatbot ChatGPT has successfully caught the attention of people with its ability to respond to queries in detail, irrespective of the data limits. And now, it has reached another major milestone in its race towards the future of Artificial Intelligence with the arrival of GPT-4.

Unlike the ChatGPT and other GPT-3.5 powered technologies, GPT-4 is expected to deliver new features like multimodality video processing and the ability to create AI-generated videos from text prompts. GPT 3.5 takes only text prompts, while GPT-4 can take image inputs to identify objects in a picture. GPT 3.5 has a limited amount of word responses (3000), while GPT-4 can generate a response of 25,000 words. OpenAI mentions GPT-4 as ‘multimodal’ as it can generate content from both text and images.GPT-4 is said to be more accurate and creative in response and has better problem-solving skills when compared to ChatGPT.

According to Sam Altman, the CEO of OpenAI, GPT-4 is the ‘most capable and aligned’ model from the company yet. “Here is GPT-4, our most capable and aligned model yet. It is available today in our API (with a waitlist) and in ChatGPT+".

Ever since these developments, people were always curious to ask the question, ‘Could AI be able to do jobs as humans do?’ Though the answer remains ambiguous, the chatbot itself has something to say. A Twitter user has asked the new GPT-4 if it can do human jobs, for which the chatbot listed 20 jobs and the human traits that humans needed to do them.

1. Data entry clerk with speed and accuracy
2. Customer service representative with communication and empathy
3. Proof-reader with attention to detail
4. Paralegal with research and organisation
5. Bookkeeper with mathematical skills
6. Translator with language proficiency
7. Copywriter with creativity and writing
8. Market research analyst with analytical skills
9. Social media manager with content creation and curation
10. Appointment scheduler with time management
11. Telemarketer with persuasion and communication
12. Virtual assistant with multitasking and organisation
13. Transcriptionist with listening and typing skills
14. News reporter with fact-checking and writing skills
15. Travel agent with planning and coordination
16. Tutor with knowledge and teaching
17. Technical support analyst with troubleshooting and problem-solving skills
18. Email marketer with writing and targeting
19. Content moderator with critical thinking and judgement
20. Recruiter with interviewing and assessment

The capabilities and limitations of GPT-4

GPT-4 has already beaten 90% of humans in the world’s toughest exam. It scored 90% in Bar Exam and 88% in LSAT. OpenAI’s president, Greg Brockman, demonstrated how GPT-3 could help people calculate their taxes which is one of the biggest use cases. The demo proved it could take a picture of a hand-drawn mock-up for a simple website and create a real one from it. Be My Eyes, an app which helps visually impaired people, is reported to provide a virtual volunteer tool powered by GPT-4 on the app. The ability to perform in low-resource languages like Latvian, Welsh and Swahili other than English is another major capability of GPT-4. 

OpenAI commented on the limitations of GPT-4, as it has few similar limitations to its older versions. They said it is less capable than humans in many real-world scenarios. Sam Altman said, “It is still flawed, limited, and seems more impressive on first use than it does after spending more time with it,”. It also lacks information about the events that occurred after September 2021, when the data was cut off. They said inaccurate responses like “hallucinations” have been another challenge for many models like GPT-4.

As the ultimate aim of these AI models is to reduce manpower and thereby increase production, advanced versions like GPT-4 will aid humans to a great extent. Irrespective of the limitations, we can definitely say GPT-4 is a significant turning point for OpenAI in their effort to scale up deep learning.