Here are 5 ways the new GPT-4 absolutely spanks the current version of ChatGPT

Here are 5 ways the new GPT-4 absolutely spanks the current version of ChatGPT. As impressive as it is, the current version of ChatGPT, which is modelled on the GPT-3 and GPT3.5 protocol, looks and feels obsolete. Sam Altman and the team of OpenAI showcased just how much more smarter and faster the new GPT-4 model is.

ChatGPT in its original form as it was released to the public, is very impressive, despite its various shortcomings as a Generative Pre-trained Transformer or a text generator. However, with the new GPT-4 update that OpenAI showcased recently, the current version of ChatGPT, which is modelled on the GPT-3 and GPT3.5 protocol, looks and feels obsolete, almost as if it was an ancient relic.

Sam Altman and the team of OpenAI showcased just how much more smarter and faster the new GPT-4 model is. Not only does it perform its operations much quicker than the regular version of ChatGPT, but, it also has a long memory and can actually follow context for longer.

No wonder that the new GPT-4 based version of ChatGPT, or ChatGPT 2.0 as some people are calling it, cleared almost every exam that it took with flying colours and got scores that would be considered in the top band of test takers.

Here are 5 ways the new GPT-4 absolutely spanks the current version of ChatGPT

Having seen how the new updated version of ChatGPT, or GhatGPT 2.0 performs, here are 5 ways we observed that the new GPT-4 model absolutely thrashes the older version.

GPT-4 “Sees” and understands images and shapes

The biggest upgrade is that GPT-4 is “multimodal,” which means it can comprehend more than one form of information. ChatGPT and GPT-3 could only communicate in text: they could read and compose.

GPT-4, on the other hand, can be given images which it then analyses to discover pertinent information. Of course, you could simply ask it to explain what’s in an image, but its comprehension extends beyond that.

Greg Brockman, an engineer, President and Co-founder of OpenAI, showed a demo, in which an updated version of ChatGPT that was running on GPT 4. It just gave it a very basic prompt, written on a napkin, and in a short time, the machine not only understood the prompt, but also wrote out a piece code that would go with the code.

A good memory and an even better attention span.

Because of costs and how much resource GPTs consume, Large Language Models are designed to have a limited scope of functionality.

Most LLMs are trained on millions of websites, webpages and other textual data. However, when in conversation with a user, earlier GPT models would have a very limited recollection of what they read, which was often represented in tokens. GPT-3 and GPT3.5 had about 4100 tokens for each conversation, or about 8000 words, after which it would just generate incoherent nonsense, out of sync with the conversation.

With GPT-4 has a maximum token count of 32,768 tokens. That’s about 64,000 words. What this means, is that GPT-4 based models, will rarely go incoherent. If we visualisea your converatoon as a book, a tool like ChatGPT will remember things fro as back as 50 pages ago with great detail.

No fooling it twice

For everything that chatbots get right today, they also get a lot of things wrong, especially when you probe them towards breaking certain rules. They are very easily misled.

A little persuasion can convince them that they are simply describing what a “bad AI” would do, or some other little fabrication that allows the model to say all kinds of strange and honestly unsettling things. People even work together to create “jailbreak” cues that let ChatGPT and other chatbots like it to go all out bonkers.

GPT-4, on the other hand, has been trained on a large number of malicious cues that broke ChatGPT last year. With these considerations in mind, the new model outperforms the previous generation in terms of “factuality, steerability, and refusal to go outside of guardrails.”

According to OpenAI, GPT-3.5, which drove ChatGPT was just a “test run” of a new training design, and the lessons learned from that were transferred to the new version, which was “unprecedentedly stable.” They were also better able to anticipate its powers, resulting in fewer surprises.

GPT-4, the linguist

Most chatbots that we have seen up until now has been very fluent in just the English language, mainly because only data sets and language models in English was used to train GPT-3.

However, the powers of large language models are relevant to any written language and should be made accessible in those as well.

Here are 5 ways the new GPT-4 absolutely spanks the current version of ChatGPT

GPT-4 makes a move in that direction by showing its ability to answer thousands of multiple-choice questions correctly in 26 languages ranging from Italian to Ukrainian to Korean. It excels at Latin and Germanic languages, but it also generalises well to others.

The many personalities of GPT-4

There is a very interesting concept in AI called “steerability.” This basically means alluding to their ability to change their behaviour on command. This can be beneficial, such as when acting as a compassionate listener, or dangerous, such as when people persuade the model that it is evil or depressed.

GPT-4 incorporates steerability more naturally than GPT-3.5, and users will be able to customise the “classic ChatGPT personality with a set verbosity, tone, and manner” to better fit their requirements. “Within limits,” the team emphasises, pointing out that this is the simplest method to get the model to break character

This could be accomplished by preparing the chatbot with instructions. But you were really just making recommendations to the “default” GPT-3.5 mentality. Developers can now bake in a point of view, conversational manner, tone, or engagement technique from the start..

Here is where we share source of knowledge (Life Hacks & Life Style ), Entertainment, Technology, Fashion, E-News and updates all over the world.

Keep on visit our website for more updates.