GPT ‘for’ AI: The news of GPT - 4

OpenAI just released GPT - 4. And the expectations are high

In a hurry? Listen to the Audio Version of the Post

Image: Lexica + Stable Diffusion

Recently, OpenAI unveiled GPT-4 on their research blog. GPT-3 was the model that was previously used by applications like ChatGPT and the Bing AI. And now it has got its upgrade.

The media was filled with buzz about the new GPT-4 way before when Microsoft CTO Andreas Braun confirmed that GPT-4 would be introduced. He said this in an event called  AI in Focus – Digital Kickoff. 

Here’s what he said at the event:- 

"We will introduce GPT-4 next week, where we have multimodal models that will offer completely different possibilities – for example, videos."

And this spread like a forest fire in the AI community. Folks on the internet pointed out that since this discussion happened at the Microsoft event, both OpenAI and Microsoft are working closer than ever before in advancing the Generative A.I. Model. The GPT-4 model is already implemented in ChatGPT+ and Bing.

The company claims the model is “more creative and collaborative than ever before” and “can solve difficult problems with greater accuracy.” GPT-4 is multimodal - meaning it can work with different types of media like images, video and others. 

The previous model GPT 3 (or GPT 3.5 since OpenAI revealed it silently) could only work with text-based data. And now with GPT-4, ChatGPT gets a boost. GPT-4 can also respond to images just like how it responded to the text-based inputs, according to the company.

OpenAI in their research blog said : 

“We’ve spent 6 months iteratively aligning GPT-4 using lessons from our adversarial testing program as well as ChatGPT, resulting in our best-ever results on factuality, steerability, and refusing to go outside of guardrail,”

So in this model (GPT-4) is expected to be much more powerful with very few inaccurate results (though in my opinion, the accuracy of this model can only be assessed after enough public testing. Right now, it seems superficial to me)

Sam Altman @sama

here is GPT-4, our most capable and aligned model yet. it is available today in our API (with a waitlist) and in ChatGPT+. openai.com/research/gpt-4 it is still flawed, still limited, and it still seems more impressive on first use than it does after you spend more time with it.

openai.comGPT-4We’ve created GPT-4, the latest milestone in OpenAI’s effort in scaling up deep learning. GPT-4 is a large multimodal model (accepting image and text inputs, emitting text outputs) that, while less capable than humans in many real-world scenarios, exhibits human-level performance on various professi…

One of the interesting aspects of GPT-4 for me is that it can now work with images. Have a look at this video: 

This video is a clip where he says about how powerful ChatGPT is and how CAN it evolve in the future. And well, minus the dramatism in the video, what he said was indeed true. Now, with GPT-4, we are closer to achieving a ‘superintelligent’ AI - stringer than ever before.

This could mean that its power could increase exponentially, accuracy would be improved, and would undoubtedly be a game changer. For instance, have a look at this tweet.

Peter Welinder @npew

Things are going to change so fast.

Mckay Wrigley @mckaywrigley

Greg Brockman (@gdb) of OpenAI just demoed GPT-4 creating a working website from an image of a sketch from his notebook. It’s the coolest thing I’ve *ever* seen in tech. If you extrapolate from that demo, the possibilities are endless. A glimpse into the future of computing. https://t.co/um6CTdJUr8 https://t.co/1QB6wbQkld

That's pretty impressive in my opinion. With such capabilities, ChatGPT would no longer be a ‘chatbot’ whose sole purpose is to chat. With GPT-4, A.I. will become a smarter ‘copilot’ for us. 

OpenAI even went further to claim that it performs better than humans on many standardised tests. Hopefully, this could remove the previously embarrassing errors by the ChatGPT. I wrote about this in past, when Arizona State University found ChatGPT was terrible at math. It scored with below 60% accuracy in the 1,000 mathematical word problems given to it by the professor. 

Right now, GPT-4 is available for ChatGPT+ users and Bing. In fact, Microsoft only recently announced that it has integrated GPT-4 in their AI search engine. This is not surprising though, since OpenAI and Microsoft are working closely together ever since Microsoft invested huge sums of money (over 10 Billion to put the right amount) to dominate the search race with its rival Google.

I will write more about GPT-4 as time goes by. For now, it would be best to go in ‘wait and watch’ mode, because when GPT-3.5 was revealed (although silently), people had the same perception - that it will become more advanced. However, the reality turned out dirty, since after its release, there was a huge spike in misinformation, toxic responses and sometimes even misleading claims. 

And this could be concerning, since ChatBots would play a major role in the future. So if anything goes terribly wrong, its effect would be massive.

So the expectations are high. Only time will tell if this disruptive technology would turn out to be good or not.

I’ll keep a watch on it and report back with more info about GPT-4. It's revolutionary for sure, but is it for good or bad? As I said, only time would tell.

See you in the next post.

OpenAI @OpenAI

Announcing GPT-4, a large multimodal model, with our best-ever results on capabilities and alignment: openai.com/product/gpt-4