Having utilized OpenAI’s latest large language model (LLM), GPT-4, since its launch, People have come to realize that access to its text-input capability is exclusively available through a ChatGPT Plus subscription, which comes at a monthly cost of $20. It must be said that for obtaining precise written responses to inquiries, this paid service surpasses its predecessor, GPT-3.5, which was the first widely available version to the public.
Let’s clarify that ChatGPT-4 is not without its flaws, occasionally referred to as “hallucinations.” However, to seek intelligent responses to complex questions, ChatGPT-4 is a valuable resource. Keeping this in perspective, let’s explore the technical distinctions between the free version of GPT-3.5 and the subscription-based GPT-4 available in ChatGPT Plus.
Distinctions between GPT-3.5 and GPT-4
- OpenAI characterizes GPT-4 as being “10 times more advanced than its predecessor, GPT-3.5.” This significant advancement empowers the model to excel in contextual comprehension and nuanced distinctions, leading to notably improved accuracy and coherence in its responses. OpenAI highlights several specific differences that contribute to this enhancement:
- OpenAI has not disclosed the exact number of parameters in GPT-4. However, according to Andrew Feldman, CEO of AI company Cerebras, GPT-4 has undergone training with approximately 100 trillion parameters, which is significantly greater than GPT-3’s 175 billion parameters.
- One remarkable feature of GPT-4 is its multimodal capability, enabling it to process both text and image data. For instance, it can receive an image as part of a prompt and deliver a suitable textual response. For example, show GPT-4 an image of the contents of the refrigerator, it can suggest recipes based on the ingredients it observes. On the other hand, ChatGPT Plus will only provide text-based answers, meaning ask for images like the Mona Lisa, it will describe similar works but won’t display any pictures.
- Another improvement in GPT-4 lies in its memory capacity. While GPT-3.5 has a short-term memory of around 8,000 words, GPT-4’s short-term memory extends to approximately 64,000 words. Additionally, GPT-4 can extract text from web pages when sharing a URL in the prompt
Advanced Features of GPT-4
- GPT-4 demonstrates enhanced multilingual capabilities, accommodating 25 languages beyond English, such as French, German, and Spanish.
- In terms of “steerability,” GPT-4 offers more control over its responses, allowing users to select a specific “personality.” For example, it can instruct it to provide answers as if it were a pirate. While the purpose behind such requests may be puzzling, GPT-4 allows for this kind of playful interaction. As ChatGPT-4 playfully phrases it, “I ain’t certain why ye’d be cravin’ to, but aye, ye can, matey!”
- Despite the limitation of GPT-4’s language model training data being confined to September 2021, it can still leverage its ability to search the internet using Bing for more up-to-date information. This feature provides the advantage of receiving more timely answers. However, it’s essential to acknowledge that this beta function is not without its imperfections.
- Introducing another beta feature known as “Plugins,” OpenAI, along with third-party developers, can enhance ChatGPT-4’s intelligence by integrating external application programming interfaces (APIs). For instance, using travel plugins like Expedia and Kayak, it can inquire about the best flights to Vancouver. Moreover, the “Ask the PDF ChatGPT Plugin” enables OpenAI to retrieve answers from any PDF document specified.
Advancements and Assessments of GPT-4
As a result of these advancements, GPT-4 surpasses its predecessor in intelligence. Notably, GPT-4 has attained higher scores on various exams, including the LSAT, SAT, Uniform Bar Exam, and GRE. OpenAI asserts that GPT-4’s answers are “40% more likely to yield factual responses compared to GPT-3.5.”
However, it should be noted that GPT-4’s intelligence is not truly remarkable. For instance, even on the AP English Language and Composition test, GPT-4 still receives a mere score of 2. Such a score would not secure admission to a preferred college, nor should it be considered sufficient for obtaining a writing-related job.
Furthermore, despite ChatGPT-4’s ability to present itself convincingly, it and other AI engines are essentially advanced auto-complete, fill-in-the-blank machines. Their responses are derived from statistically probable words most likely to emerge from their language models.
According to OpenAI, GPT-4 boasts an 82% reduction in the likelihood of responding to prohibited prompts and a 60% decrease in fabricating facts.
Considering these improvements, GPT-4 appears to be a better option, and investing in ChatGPT Plus seems justified. However, it’s essential to recognize that while ChatGPT-4 offers valuable assistance, it should not be seen as a substitute for human workers but rather a helpful tool for work purposes.