GPT-4: How to access, ways it outsmarts ChatGPT and what it means for AI chatbots
OpenAI today launched GPT-4, its next-generation flagship model that forms the technical foundation for Microsoft’s ChatGPT and Bing AI chatbots. This is a profound improvement, potentially opening the door to major advances in the capabilities and performance of today’s AI technology.
How can I access GPT-4 for ChatGPT+?
If you want to generate free text for the GPT-4 model, you will need a premium subscription to ChatGPT+, the premium version of ChatGPT. Here’s how to get it:
As OpenAI’s announcement page for GPT-4 makes clear, GPT-4 is a model used by ChatGPT, but only if you are a paid member with access to ChatGPT+ (sometimes referred to as “ChatGPT Plus”). If you’re logged into your OpenAI account, the button on this page will take you directly to ChatGPT, and if you’re a paid ChatGPT+ user, that’s it. Have fun!
If, however, you are using a free subscription, you will need to upgrade. Click the “Upgrade to Add-on” button on the left menu.
A pop-up window will appear, informing you about the benefits of the “Plus” upgrade.
The green “Upgrade Plan” button will take you to the usual e-commerce sales page, where you have to enter the usual information and cough up $20 a month.
Once you become a paying customer, your access to the new style through ChatGPT will be immediate. There is no waiting list for you.
When you start a new chat, a drop-down menu will give you the option to use either the old format or GPT-4. You’ll want to heed OpenAI’s warning that GPT-4 isn’t as fast as others, as the speed difference is huge.
If you are already using ChatGPT, you may be prompted to upgrade to GPT-4. If you do, a text bar will be added to the middle of your existing chat, saying, “The previous template used in this chat is unavailable. We’ve redirected you to the newer default template.” Legacy versions that used GPT-3 will have the old green OpenAI logo to the left of them, while versions running GPT-4 will have the new black logo.
How to get GPT-4 for free
Currently you cannot access GPT-4 with the free version of ChatGPT.
However, millions of Bing users already have access to the new Bing chatbot. If you fall into this chosen category, Microsoft has confirmed that GPT-4 already enables your interaction with the chatbot.
And soon, you may have no choice but to use GPT-4 in your daily life, because according to OpenAI’s blog post about its use, GPT-4 is already being used by different clients like Khan Academy, Stripe. , Morgan Stanley and the Icelandic government. In other words, there’s a good chance you’ll be dealing with GPT-4 soon, whether you like it or not.
the difference between the chatbot you know and love and its newly added successor.
- GPT-4 can see and understand images
The most notable change to this multi-machine learning system is that it is “multimodal”, meaning it can understand more than one “module” of information. ChatGPT and GPT-3 were limited to text: they could read and write, but that was about it (although it was more than enough for most applications).
GPT-4, however, can receive images and will process them to obtain useful information. You can just ask him to describe what is in the picture, of course, but more importantly, his understanding goes beyond that. The example provided by OpenAI actually tells a joke in the image of an impressively sized iPhone connector, but a collaboration with Be My Eyes, an app used by the blind and visually impaired to allow volunteers to describe what their phone sees, reveals more. .
In the Be My Eyes video, GPT-4 explains the design of a dress, identifies a machine, explains how to get to a certain machine at the gym, interprets a label (and gives a recipe), reads a map, and does some numbers. of other works that really show he gets what’s in the picture – if he’s asked the right questions. He knows what the outfit looks like, but he may not know if it’s the right outfit for your interview.
- GPT-4 is difficult to cheat
Despite all that today’s chatbots achieve, they are easily misled. A little coaxing might convince them that they’re just explaining what a “bad AI” would do, or some other little story that leaves the model saying all sorts of weird and downright annoying things. People even collaborate on “jail crisis” tips that quickly release ChatGPT and others from their confinement.
GPT-4, on the other hand, was trained on many malicious tips – which users have provided to OpenAI over the past several years. Considering these factors, the new model is better than its predecessors on “truth, cunning and rejection from the guards”.
As OpenAI describes it, GPT-3.5 (which ran ChatGPT) was a “test” of the new training architecture, and they applied the lessons to the new version, which was “very stable”. They also had a better ability to predict his abilities, which reduces surprises.
- GPT-4 has long memory
These huge language models have been trained on millions of web pages, books, and other textual data, but when they have a conversation with a user, there is a limit to what they can keep “live”. ‘spirit’, so to speak (we sympathize). This limit for GPT-3.5 and the older version of ChatGPT was 4,096 “tokens”, or about 8,000 words, or about four to five pages of a book. So it would lose track of things after going far “backward” in its focus function.
GPT-4 has a maximum token count of 32,768 – that’s 2^15, if you’re wondering why that number seems odd. That translates to about 64,000 words or 50 pages of text, enough for an entire play or short story.
This means that in a conversation or when making a text, it will be able to remember up to 50 pages. So he will remember what you said about 20 pages of conversation back, or, when writing a story or an essay, he can refer to events that happened 35 pages ago. This is a very rough description of how the attention mechanism works and the symbolic computation, but the general idea is one of extended memory and the capabilities that come with it.
- GPT-4 has more languages
The world of AI is dominated by English speakers, and everything from data to experiments to research papers is in that language. But of course, the power of the main language structures is used in any written language and should be found in it.
GPT-4 takes a step in this direction by demonstrating that it can answer thousands of multiple-choice questions with high accuracy in 26 languages, from Italian to Ukrainian to Korean. It is best in Romance and German languages but generalizes well to others.