After months of rumors and reports, OpenAI has finally unveiled the new ChatGPT 4 claiming it to be “more creative and collaborative than ever before.” The new language model has outperformed its sibling ChatGPT by scoring in higher approximate percentiles among test-takers for Uniform Bar Exam.
OpenAI based on its internal findings claims that the new ChatGPT 4 is “82% less likely to respond to requests for disallowed content and 40% more likely to produce factual responses than GPT-3.5.” The company has been internally testing GPT-4 focussing on functions like support, sales, content moderation, and programming.
Why they launched GPT-4 so early?
This new model is deemed to be an upgrade in almost every aspect. OpenAI claims it has the ability to offer "human-level performance" and can solve difficult problems with greater accuracy compared to its predecessor, GPT-3.5, which powers the ChatGPT. This is due to its broader general knowledge and problem-solving abilities.
The new capabilities of GPT-4 are expected to revolutionize the conversational AI industry and enhance the user experience. However, to check how big of an upgrade it really is, we tried to pitch the standard version of ChatGPT with Bing AI Chat which is using GPT-4. However, it is noteworthy that Microsoft has customised the Bing AI chat for the search engine.
Bing AI Chat vs ChatGPT
Bing AI Chat is primarily designed to provide quick, relevant answers to specific queries, ChatGPT on the other hand is a large language model trained to provide detailed and informative responses to a wide range of topics.
The main difference between the two is in their scope of information and the level of detail they provide. Bing AI Chat offers simple, straightforward responses to direct queries, while ChatGPT provides more in-depth and comprehensive answers.
When both language models were quizzed about the best Biryani joints in Delhi, they surprisingly offered similar suggestions. Bing AI Chat extracts real-time data from the internet but ChatGPT still works on dated data. Despite that, ChatGPT managed to keep up with GPT-4 based Bing Chat.
Overall, while both Bing AI Chat and ChatGPT are valuable tools for accessing information, they offer different levels of detail and depth, making them suited for different purposes and types of queries.
Strength & Weakness Comparison of Bing chat and GPT-4
- Chatgpt has a very large and diverse knowledge base, as it is trained on a huge corpus of text from the internet. Chatgpt can chat about almost any topic and provide relevant and interesting information, facts, opinions, or trivia.
- Chatgpt can also adapt to the user's style, tone, and personality, and can generate responses that are natural, engaging, and sometimes humorous. Chatgpt can also handle multiple turns of conversation and maintain coherence and consistency.
- Bing chat has a very powerful and accurate web search engine, as it is backed by Microsoft Bing search. Bing chat can perform web searches when the user is seeking information or whenever search results could be potentially helpful, regardless of Bing chat's internal knowledge or information.
- Bing chat can also provide informative, visual, logical and actionable responses to user messages, and can also generate creative content such as poems, stories, code, essays, songs, celebrity parodies and more. Bing chat can also leverage information from multiple search results to respond comprehensively and thoroughly.
- Chatgpt has some limitations in terms of its accuracy, reliability, and safety. Chatgpt may sometimes generate responses that are inaccurate, irrelevant, or inappropriate, as it is not always able to verify the information or the context. Chatgpt may also sometimes generate responses that are harmful to someone physically, emotionally, financially, or morally, as it is not always able to detect the user's intent or the ethical implications.
- Bing chat has some limitations in terms of its flexibility, creativity, and personality. Bing chat may sometimes generate responses that are too factual, dry, or boring, as it is not always able to generate responses that are natural, engaging, or humorous. Bing chat may also sometimes generate responses that are too generic, vague, or off-topic, as it is not always able to adapt to the user's style, tone, and personality.
OpenAI official statements
OpenAI says that the GPT-4 is more reliable, creative, and able to handle much more nuanced instructions than GPT-3.5. However, the company also cautions that the new model may also show similar problems as the previous one like making up information and generate harmful text.
New features of GPT 4
- One of the key features of GPT-4 is its ability to accept images as inputs and generate captions, classifications, and analyses.
- Users can share images with GPT-4 and receive responses according to the visuals.
- Additionally, GPT-4 can handle much longer text compared to its predecessor, with the ability to use and analyze over 25,000 words of text.
While the latest ChatGPT 4 has also not been trained on the recent event, company says that it can solve difficult problems with higher accuracy. The chatbot has been trained on more human feedback, including feedback submitted by ChatGPT users, to improve its behaviour. The company informs that it has worked with over 50 experts for early feedback in domains including AI safety and security.