People were amazed by the natural language abilities of ChatGPT, an AI chatbot that took the world by storm. However, the arrival of GPT-4, a massive language model, shattered all expectations. Some even consider it the early glimpse of artificial general intelligence (AGI). Let’s dive into everything you need to know about GPT-4, including how to use it and its remarkable capabilities.
Availability
Image source: Shutterstock
GPT-4 was officially introduced on March 13, as previously confirmed by Microsoft. Presently, it’s accessible exclusively through the ChatGPT Plus paid subscription. The free version of ChatGPT is still based on GPT-3.5, which is less accurate and capable compared to GPT-4.
Furthermore, GPT-4 has been made available as an API for developers to create applications and services. Several companies, including Duolingo, Be My Eyes, Stripe, and Khan Academy, have already integrated GPT-4 into their systems. The first public demonstration of GPT-4 was also livestreamed on YouTube, showcasing its new capabilities.
What’s New in GPT-4?
GPT-4 is a groundbreaking language model developed by OpenAI that can generate text similar to human speech. It improves upon the technology used in ChatGPT, which is currently based on GPT-3.5. GPT stands for Generative Pre-trained Transformer, a deep learning technology that mimics human writing using artificial neural networks.
OpenAI claims that GPT-4 outperforms ChatGPT in three key areas: creativity, visual input, and longer context. In terms of creativity, GPT-4 excels at generating and collaborating on creative projects with users. These projects include music, screenplays, technical writing, and even learning a user’s writing style.
Longer context is another area where GPT-4 shines. It can now process up to 25,000 words of text from the user, and users can even provide GPT-4 with a web link and ask it to interact with the content on that page. This feature is particularly useful for creating long-form content and engaging in extended conversations.
Additionally, GPT-4 can now receive images as input for interaction. For instance, it can be given an image of ingredients and asked what can be made with them. However, it’s unclear whether videos can be used in the same way.
Image used with permission by copyright holder
Lastly, OpenAI emphasizes that GPT-4 is significantly safer to use compared to its predecessor. In internal testing, it reportedly produced 40% more factual responses and was 82% less likely to respond to requests for disallowed content. OpenAI achieved these improvements through extensive training with human feedback, collaborating with over 50 experts in AI safety and security.
Users have already showcased incredible achievements with GPT-4, such as inventing new languages, creating detailed real-world escape plans, and developing complex animations for apps from scratch. As early adopters experiment with the technology, we are beginning to uncover its vast potential.
Is GPT-4 Experiencing Decline?
Despite the initial excitement surrounding GPT-4, some users have noticed a decline in the quality of its responses over time. This observation has been shared by prominent figures in the developer community and discussed on OpenAI’s forums. Initially, OpenAI dismissed these claims, but a study was later published that confirmed the degradation in answer quality. By comparing GPT-4 between March and June, researchers found a significant drop in accuracy, from 97.6% to 2.4%.
While this study is not definitive proof, it does lend support to the concerns raised by users. It appears that the decline is not merely imagined.
Visual Input in GPT-4
One of the most anticipated features of GPT-4 is its ability to process visual input, enabling ChatGPT Plus to interact with images rather than just text. However, this feature has been delayed due to safety challenges, according to OpenAI CEO Sam Altman.
You can catch a glimpse of visual input in action in Bing Chat, where the feature has been recently introduced for select users. Additionally, an open-source project called MiniGPT-4 allows users to experiment with visual input. Although image processing may be slower in this project, it provides a preview of the tasks that will be available when visual input is officially rolled out for GPT-4 in ChatGPT Plus.
The Best GPT-4 Plugins
Image used with permission by copyright holder
Plugins greatly enhance the capabilities of GPT-4 and are one of the key advantages of subscribing to ChatGPT Plus. OpenAI has developed two powerful plugins: the code interpreter and web browser plugins.
The ChatGPT Code Interpreter plugin offers a persistent session where you can use Python, including handling uploads and downloads. On the other hand, the web browser plugin grants GPT-4 access to the entire internet. This plugin allows GPT-4 to bypass model limitations and retrieve live information from the web on your behalf.
Other notable GPT-4 plugins include Zapier, Wolfram, and Speak, all of which unlock new possibilities for AI.
How to Use GPT-4
Image by Jacob Roach / Digital Trends
Currently, the primary way to access GPT-4 is by upgrading to ChatGPT Plus. To upgrade your subscription to the $20 paid plan, simply click on “Upgrade to Plus” in the ChatGPT sidebar. After entering your credit card details, you will be able to switch between GPT-4 and older versions of the language model. GPT-4 responses can be identified by a black logo instead of the green logo used for older models.
Once you have access to GPT-4, using it is identical to using ChatGPT Plus with GPT-3.5. GPT-4 offers greater capabilities than ChatGPT, allowing you to fine-tune datasets for tailored results that meet your specific requirements.
If you prefer not to pay, there are alternative ways to experience the power of GPT-4. For example, Microsoft’s Bing Chat utilizes GPT-4 and is completely free to use. However, some GPT-4 features might not be available in Bing Chat, as it incorporates aspects of Microsoft’s proprietary technology. Another option is to visit question-answering sites like Quora, which currently utilize GPT-4.
Limitations of GPT-4
While discussing the new capabilities of GPT-4, OpenAI acknowledges certain limitations of the language model. Like its predecessors, GPT-4 still grapples with issues such as “social biases, hallucinations, and adversarial prompts.”
In other words, GPT-4 is not perfect. It may occasionally provide incorrect answers, and its limitations have been documented online. However, OpenAI is actively working to address these issues, and, in general, GPT-4 is “less creative” and less likely to generate fictitious information.
Another key limitation is that GPT-4 was trained on internet data only up until 2021. The web browser plugin is a workaround to mitigate this limitation.
Does Bing Chat Use GPT-4?
Image used with permission by copyright holder
Initially, Microsoft hinted that its new product, Bing Chat, was more powerful than ChatGPT. Given that OpenAI’s chat utilized GPT-3.5, it led to speculation that Bing Chat might be driven by GPT-4. Eventually, Microsoft confirmed that Bing Chat is indeed built on GPT-4.
However, features like visual input were not available in Bing Chat, so it remains unclear which exact features have been integrated and which have not.
Regardless, Bing Chat represents a significant upgrade, enabling access to current information from the internet, which surpasses the capabilities of the current version of ChatGPT, limited to the training it received up until 2021. Moreover, the AI model used in Bing Chat is much faster, a crucial factor when applied to a search engine.
An Evolutionary Step
While we haven’t personally tested GPT-4 in ChatGPT Plus, it is undoubtedly expected to be more impressive, building upon the success of ChatGPT. In a StrictlyVC interview on YouTube, OpenAI CEO Sam Altman warned that “people are begging to be disappointed, and they will be.”
Altman emphasized the potential of AGI to disrupt global economies. He expressed that a gradual rollout of several incremental changes is more advantageous than sudden advancements that leave the world struggling to adapt. Though GPT-4 appears impressive, it represents a careful evolution rather than a complete revolution.
Editors’ Recommendations
- ChatGPT: The Latest News, Controversies, and Tips You Need to Know
- Google Bard could soon become your new AI life coach
- Zoom backpedals, says it will no longer use user content to train AI
- Newegg wants you to trust ChatGPT for product reviews
- AI can now steal your passwords with almost 100% accuracy — here’s how
Note: The above links are only accessible within the context of this article for “OnSpec Electronic, Inc.”
