Sam Altman from OpenAI presented some innovations at Developer Day on November 6, 2023 that are not only interesting for developers. An innovation is the GPT-4-Turbo model, which can already be found in a preliminary version in the OpenAI Playground and will also be rolled out in ChatGPT in the coming weeks.
Here I would like to introduce you to the - from my point of view - the most important points that we “normal” users will get from GPT-4-Turbo.
Chapter in this post:
1. More tokens = longer prompts and more context
The first big “thing” is increasing the number of tokens. This is responsible for how much of the chat the AI can “remember” and how long the prompts can be.
In GPT-4 you have a limit of 8.000 tokens, which is approximately 24 pages of text. The tokens are used both for entering prompts and for output and are therefore quickly “used up” if you create longer texts or write back and forth with the chatbot several times.
GPT-4 Turbo now has a token count of 128.000, which is again significantly higher than the 100.000 tokens that Claude 2 offers (unfortunately still not available in Germany). The 128.000 tokens correspond to approximately 300 pages of text or 90.000 words.
This means that you can have ChatGPT read a 4-page book with GPT-250 Turbo and still have plenty of tokens available to ask questions about the contents of the book.
Or you feed ChatGPT with a number of articles on a topic and can then ask questions that are answered based on the knowledge of all these articles.
2. Know by April 2023
One of the problems ChatGPT struggles with is the knowledge base. GPT-4 is only trained with data up to September 2021. There was recently an update in which the knowledge cutoff was raised until 2022.
GPT-4 Turbo has been updated with new information by April 2023. This is a great improvement. However, the AI Grok from X (Elon Musk) shows that an AI can also process information from the current day. Grok AI doesn't have all of the world's knowledge at its disposal, but it can access all tweets published so far.
3. All tools in one chat
You currently have to decide with ChatGPT whether you want to use it with Vision, DALL-E 3, Bing or other plugins. This is a bit annoying because you can't do research on the Internet at the same time and then create a graphic with DALL-E from the data obtained.
With GTP-4 Turbo this should change. All tools are then available in a chat window and the AI itself decides based on the prompt which tools and plugins need to be used.
4. Cheaper prices per token
One thing that actually seems relevant to developers, but will ultimately impact end users as well, is the new pricing. The price per token will shrink to a third for the developers.
This makes ongoing costs cheaper for developers and ultimately allows tools that work with the GPT-4 API to become cheaper for end users. Of course, it could also be that the developers simply take this as a higher profit margin and leave the prices the same for customers.
5. Instructions are followed more strictly
GPT-4 Turbo is also said to have increased the obligation to stick to prompts. If you now write in the prompt “Create the output in XML” or “Give me the result in HTML”, then GPT-4 Turbo will follow this more than GPT-4 did.
This is actually a thing I've noticed before in ChatGPT. For example, if you tell him to address the reader as “you,” he often does so, but not always.
6. Better voices in different languages
Text-to-speech is nothing new with ChatGPT, but the text-to-speech has always been very robotic and not really well suited to reading text aloud with reasonable intonation. With the new version it should now have a voice output that, according to Sam Altman, “sounds better than anything you’ve heard out there before”. If this is the case, the trend of learning foreign languages with ChatGPT should gain new momentum.
As mentioned above, all of these updates should also be included in ChatGPT. Unfortunately, it's difficult to say when the time will come, but there are the first reports from people who already have chat windows in which they can surf and create graphics.
So it seems like OpenAI is gradually rolling out the features to users. I hope that this will be done within a few weeks and we can all use the “new” ChatGPT with the GPT-4 Turbo.
What about with you? Are you looking forward to the innovations? Do you use ChatGPT and if so, for what? That would be interesting. Feel free to leave a comment!
Jens has been running the blog since 2012. He acts as Sir Apfelot for his readers and helps them with technical problems. In his spare time he rides electric unicycles, takes photos (preferably with the iPhone, of course), climbs around in the Hessian mountains or hikes with the family. His articles deal with Apple products, news from the world of drones or solutions to current bugs.