The release of the GPT-4 brought with it an incredible wave of excitement, dismay, a range of emotions between the two poles, and a fundamental question. When will the GPT-5 come out?
For now, we don’t know much, so this entire article is pieced together from speculation, conjecture, and bits and pieces of information that have surfaced in the public domain.
Will the GPT-5 ever come out?
The GPT-5 will most likely come out, although it’s not quite on the agenda at the moment. OpenAI CEO Sam Altman stated bluntly during his interview with The Economic Times person that “the systems we have now are not important at all. In time, we’ll look back at how imperfect GPT-4 Chat is and say, “We were so excited about this?”, it’s like picking up the first iPhone that came out today.”
Furthermore, Sam mentioned in the interview that we’re at the beginning of an exponential growth curve because of the OpenAI algorithm:
Learn and improve in the sense that we understand those words.
Improving as the number of its parameters grows.
From these words, then, I deduce that we are in for not only Chat GPT-5, but GPT-6, GPT-7, GPT-8… and beyond.
When will GPT-5 be released?
Based on previous GPT model releases and the current tension around AI, it is safe to guess that GPT-5 will not be released in 2023, as the GPT-4.5 model, which is likely to precede the development of GPT-5, is not yet on the agenda.
Some experts have suggested that OpenAI could introduce an intermediate GPT-4.5 model as early as October 2023, similar to what happened with GPT-3.5, but we will have to wait to see if this is speculation or if it will actually happen.
What will the GPT-4.5 intermediate model bring?
It is expected that GPT-4.5 could bring the long-awaited multimodal capability that will allow analysis of both images and text -> which is already possible in the paid version using the Avanced data analysis tool, formerly GPT Code Interpreter.
However, the features and speed of development will certainly depend to a large extent on the company’s ability to allay the public’s concerns and criticisms that OpenAI is still quite subject to. And personally, I think that GPT-4.5 will be a gradual improvement rather than a complete overhaul.
What are the expected features of GPT-5?
Below you will find a basic list of points where I expect the model to be better. At the same time, it is also safe to say that improvements on these points are likely to be seen in any future version of Chat-GPT.
- Multimodal capabilities GPT-5 should be able to handle a variety of data types including text, images, video and audio. Compared to GPT-4, this is the biggest change and could open up completely new uses and ways of interacting with the model.
- Factual accuracy and less hallucination GPT-5 will aim to be more factually accurate and less hallucinating. Given the probabilistic nature of large language models, inaccuracies and hallucinations naturally occur frequently. And since the model will give a different answer to the same question every time, the creators’ desire to increase accuracy is only logical.
- Long-term memory and improved context retention With the trend of increasing the number of characters your message to GPT can contain, the trend for the model to retain the context of your conversation better and further goes hand in hand. GPT-5, in my opinion then, will definitely only continue this trend.
- Expanded Knowledge Base As with any new release, GPT-5 is expected to have a broader understanding of various areas, including current events, scientific advances, and popular culture. The success and potential of this, then, depends, in my opinion, primarily on the progress of the legislation and whether the GPT-5 team will have the opportunity to have the model trained on a corpus of current online data.
User customization and fine-tuning GPT-5 should again be a bit more personalized and responsive to individual user needs compared to its predecessor.
Conclusion
While we don’t know exactly what GPT-5 will be able to do, when it will be released, or what it will bring, one thing is certain: large language models have changed the world as we know it, and it will be up to us to decide how we deal with it.
I still see a lot of people fixating on the partial imperfections of the models, arguing that since GPT can’t do X or Y, it’s not really useful (at least to them). While these people are right that the models can’t do many things yet, they forget that we’re on an exponential curve of evolution, and the models will (perhaps sooner than we think) be able to do these things.
And from my point of view, it is a very good idea to take this into account, prepare for it, and gradually make friends with models and their use in everyday life.
I will do my best on this blog to help you with this task).
???? And let me know in the comments how you see it with GPT-5 and the development of language models?