Here it comes: Everything new in ChatGPT4

More creative, smarter, provides more detailed answers and can also decode images: OpenAI’s smart chat bot is getting an upgrade and offers you a power and brain multiplier of another order of magnitude

If you’ve been on Earth in recent months, you’ve probably heard of, and maybe even experienced, ChatGPT. OpenAI’s advanced bot has shaken up the world, no less, with quite a few sectors, from education and academia to banking and cybersecurity, racking their brains over how to digest it. And as often happens with new technologies, some are already calculating the end of their profession backwards.

A new version of the language modeling software, GPT4, was recently launched, and with it ChatGPT Plus, which is offered on its basis for $20 per month (mainly as a tool for programmers via an API). OpenAI promised, even against the criticism that was heard about the chat’s mistakes or evasive answers, that this version is more accurate, more creative and much more communicative, to the point of “human-level performance”.

On the surface, it is more of an evolution than a revolution. Still, the tool has a number of important innovations that are going to make life even easier.

So what’s new in the current version?

For starters, it addresses not only texts – but also visuals. This means, in general, that if you show it a picture next to text, it will be able to process both and identify connections between them. The answers will be provided verbatim in any case. This is still convenient, for example, if you ask the software for a recipe for a quick meal based on pictures of random ingredients you found in the kitchen. The company promises that similar analysis of video files is already within reach.

As mentioned, the new version is more creative and excels in various and varied types of writing, from professional-technical text to poems and plays. While working together with the author, it learns his writing style and applies it itself.

Another difference concerns the level of detail in the answers: compared to the 3,000-word limit we knew in the previous version, GPT4 can provide up to 25,000 words in each answer. So if you were building it to write a reasoned article or a short book on a professional subject, you can start talking about that.

The developer claims that in the new version, the chance of receiving reliable, well-founded, and accurate information in response to a query is 40% higher than in the previous version. For example, in the US Bar Association exams, the bot managed to score in the top ten. Another promise is an 82% improvement in terms of unauthorized requests – from offensive statements to requests for malicious code.

What’s still missing?

A significant portion of the criticism of the software so far has concerned current events, with several embarrassing answers received by users. The reason for this is simple: the software does not really capture what has happened in the world since September 2021. Admittedly, it continues to learn on the fly, but certainly not on the scale of intensive algorithm training and database input for the formatting stages.

Current events aside, it seems that the level of reliability itself is still lacking. Issues related to privacy, offensive discourse, social biases, and the spread of prejudice also remain in need of improvement. And for those who use the software for programming, according to initial tests, even the latest version may write code that is relatively easy to hack.

Another weak point will disappoint many Israelis: the company did mention that they added broader support for foreign languages ​​(not English), but in the meantime, even this version is not enough for complex tasks in Hebrew.

Where do we go from here?

The launch of ChatGPT in December 2022 stunned the world and accelerated the arms race of tech giants combined with artificial intelligence. Microsoft, which took the lead with a smart investment in OpenAI’s chatbot back in 2019, added another $10 billion in January of this year, on its way to a surprising life-saving injection for its Bing search engine. In February, Google launched Bard, a competitor to GPT. It is based on the LaMDA language model it developed and is set to be integrated into its popular search engine.

It is still too early to assess the scale of the revolution from the mere fact that generative AI applications are accessible to everyone and for free. In the worlds of content creation, advertising and digital marketing, this is already a real game changer.