Have you noticed that ChatGPT only seems to know about events up to 2021?
In fact, when you ask it, ChatGPT will directly tell you that it only knows about events up to September 2021.
So, will OpenAI ever update ChatGPT to 2023, or even beyond? Or, will it forever be stuck in the past?
In this article, I’ll cover everything you need to know. I’ll also share a neat trick you can use to get OpenAI’s GPT-4 model to talk about current events as well.
So, let’s get started!
Will ChatGPT Be Updated To The Current Date?
At this time, it seems unlikely that OpenAI has any intentions of updating ChatGPT to the current date, or even up to 2023 at all.
While ChatGPT may know a limited amount of information about current events, this amount is extremely low compared to what it knows up to September 2021. I’ll talk more about why this is the case a little later, but as of right now, no plans have been announced to update ChatGPT.
If you think about it, it makes sense.
ChatGPT is a free product, and is costing OpenAI a lot of money to run, strictly for research purposes. By the same token, they already have more powerful language models available (such as GPT-4, with GPT-5 being actively worked on) that are more current than ChatGPT is.
The vast majority of queries do not rely on knowledge after 2021 anyway, so there is little incentive to update it. Again, especially for a free product.
That being said, OpenAI has pushed small updates out to ChatGPT – they just haven’t fully retrained it on current data.
So, there may be some cases where ChatGPT seems to understand events that happened after September 2021, even though this would be after its training cutoff date.
Alternative: Use Bing Chat
Did you know that there is as language model even more powerful than ChatGPT, that you can use right now for free?
It’s called Bing Chat, and it runs inside of the Bing search engine.
Bing Chat runs on an updated version of one of OpenAI’s models, called GPT-4. ChatGPT on the other hand can sort of be thought of as ‘GPT 3.5’ – it is still a huge upgrade over GPT-3, but GPT-4 is even more powerful.
What makes it special in this case however is that Bing Chat is connected to the internet, and is able to actively pull and analyze new information that you give it.
In fact, you can even show it specific web pages, and ask it questions based off of those pages.
Pretty neat!
This effectively means it’s always up-to-date for certain use cases, even if OpenAI isn’t actively training it on new data.
If you aren’t sure how models like ChatGPT are trained (and hence why they are limited to certain dates), keep reading – I’ll explain it in just a moment!
Why Is ChatGPT Data Limited After 2021?
In order to understand why ChatGPT only goes up to September 2021, it’s important to know a bit about how it was created at the first place.
Many people are under the misconception that ChatGPT is still learning from its users, or that it’s still scraping the internet for new information to feed itself. Neither of these are true.
Instead, ChatGPT was trained all at once using extremely large amounts of data – around 570 gigabytes of text to be exact, consisting of everything from books, to forum discussions, to articles that have been published online.
Essentially, you can kind of imagine someone reading over billions of pages, memorizing every single thing perfectly, and then using all of that information to answer questions about well, anything (since information has been published on just about everything.)
The problem was, all this data OpenAI trained on only went up to September 2021.
If they were to update it to 2023, they would need to retrain ChatGPT on all of that new data that has been published since then.
Training large language models like ChatGPT is extremely expensive. In fact, it’s estimated that ChatGPT itself would cost millions of dollars to retrain, making it quite costly to do so.
Although OpenAI certainly has the funds to do so, its primary focus seems to be on future models, rather than updating ChatGPT.
OpenAI Is Also Focusing On Other Models
This is the big reason why we’re unlikely to see ChatGPT get updated.
OpenAI has already moved beyond ChatGPT with its release of GPT-4 in March, 2023. GPT-4 is not only far more powerful than ChatGPT, but further aligns with OpenAI’s goals – namely, it is better at rejecting content that goes against OpenAI’s policies, and is better at delivering answers that are factually correct.
It also boasts a much larger context size (which can be thought of like a short-term memory) and overall yields higher quality answers.
But even if that weren’t enough, OpenAI is already working on the next version as well, GPT-5.
Although ChatGPT may still be the most popular free model open to the public, it is far from the most powerful model that OpenAI has to offer – or will have to offer in the near future.
Keep in mind that OpenAI aren’t the only players in the AI race, however.
Google has released Bard, Facebook has released LLaMA for research purposes, and other companies are continuing to release models that are more up-to-date than ChatGPT as well.
So, we’re likely to have a lot more options available to us in the near future as these models continue to roll out!
Conclusion
Currently, ChatGPT is only trained up to September 2021. While it may seem to have some understanding of current events, this information will be extremely limited compared to information before 9/21.
Because OpenAI is focused on GPT-4 and GPT-5, it is unlikely we’ll see any drastic training updates to ChatGPT in the near future. However, it is not outside of the realm of possibility either.
Thankfully, even with its limited knowledge of 2022, 2023 and beyond, ChatGPT is still more than powerful enough for most queries. And if you need something even more powerful, simply use Bing Chat for free!
I hope that this article has answered your questions. If you have any other questions about ChatGPT, ask them below and I’ll get back to you as soon as I can.
Wishing you the best,
– James McAllister