Artificial Intelligence: What Is OpenAi Chat GPT Version 4

Chat GPT version 4 has finally arrived. It leaves the chat GPT in the dust. It can convert a drawing on a napkin into a functional website it can even explain to you a joke from a series of images. So let’s find out exactly what Chat GPT version 4 actually is. Open AI ran a developer live stream that not only introduced this multimodal AI they also ran a demo to demonstrate how it’s one of the most powerful artificial intelligence engines to this day.

It’s no surprise that the whole internet Twitter especially has gone insane hearing this news chat GPT was only released recently with Bing AI. following in this post I’m going to cover what Chat GPT version 4 actually is and how it’s different from the previous GPT versions such as Chat Gpt3 and GPT 3.5 and I’ve also gained access to test it out. So I’ll demo some of its use cases too.

Chat GPT version 4: Comparison To Its Previous Models

Let’s start off with Chat GPT version 4 in comparison to its previous models, it’s multimodal now which simply means that unlike previous versions of GPT which were only text-based. Chat Gpt version 4 can accept and process images as well as text.

I wouldn’t be surprised if there’s an audio component to this as well, but for the time being open AI simply did a demonstration showcasing a few examples of their image-to-text processing, the first one is on their website where they gave AI the difficult task of explaining a joke based on a series of images.

This image is of an iPhone, but it’s charging with a VGA cable, and Chat GPT version 4 accurately was able to identify all the elements in the photo as well as explain the context of the joke while this sounds pretty simple. This is something that’s almost unheard of in previous versions of AI and I guess that Recaptcha will now have a run for its money trying to detect and not just Bots, but artificial intelligence.

The second demo is where the real magic happened. Open AI showed how you could draw a picture of a website on a napkin or maybe a piece of paper, then you can take a photo of it and send it to Chat GPT version 4. You can ask it to produce a functional website.

In this case, it was a joke’s website in just 10 to 20 seconds Chat GPT version 4 produced all the HTML CSS and JavaScript code to reproduce the website and then it was copy-pasted into an editor and showcased as a functional product, as a developer, I have to say I’m very impressed with this is something not very easily done.

Openai has also showcased some of the companies that are already working with them to be able to use Chat GPT version 4 in their products. These companies have integrated it into Services One of them is Khan Academy, they’ve integrated it very similar to Chat GPT, but with more customization to work as a personal tutor for those people who are learning educational content.

Maybe in the future, all our children will be taught by AI but at least for the short term, it looks like it’s definitely a great assistant to have on hand whether you’re doing any type of learning. There were lots of Statistics that showcased how Chat GPT version 4 performs better than any other model today.

Including its Bing ability to pass the Lstat and the bar and Bing in the top quarter percentile whereas previous versions of gpt3 were in the lower quarter of that percentile. Other than the difference Bing is able to also take in visual inputs it can also produce and handle over 25,000 words of text which is much larger than previous models.

Capabilities In GPT-4

It’s also much more creative Bing able to edit as well as modify and iterate over technical tasks and writing tasks way more accurately than previous models For example, you could ask chat GPT or gpt3 to summarize Cinderella and while it could do so it wouldn’t be able to do complex tasks such as being able to summarize it where every sentence of each word begins with the next letter of the alphabet A to Z.

This is a much more complex task and yet, a Chat GPT version 4 can do this quite easily gpt4 also surpasses Chat GPT in advanced reasoning capabilities which means that if you’re trying to book an appointment between two people’s calendars with different availabilities, gpt4 can better reason and figure out a time that works for both of them. It’s also safer and less prone to making errors. OpenAi said,

“They spent six months making sure that gpt4 is 82 percent less likely to create requests for disallowed content and 40 less likely to produce fake news or at least factually inaccurate responses.”

If you’re interested in using it right now you can do so on Chat GPT Plus which is the paid version of Chat GPT and if you want to get access to the API you’ll need to join the API waitlist. I went on to the Chat GPT website where I got the new model available Chat GPT version 4.

It also showcases the difference between each model with the differences showing in reasoning speed and conciseness. Version 3.5 has average reasoning and low conciseness but quite high speed. The Legacy version is less used and its speed is a bit lower but, Chat Gpt4 has very high reasoning and high conciseness.

But the speed is a little bit lower. I’d say it’s because it’s still being pushed down so it’s currently limited to 100 messages every four hours. I first asked if it showcases three different things that Chat GPT version 4 can do Chat GPT3 couldn’t technically be trained on the same data all the way up to September 2011.

What I found surprising is that it still thinks it’s a version of gpt3 which maybe it was trained against, but it did have the correct answers which means that it is better at comprehension and understanding.

It is better at reasoning and it’s also better as a language model supporting more languages more accurately well in my last post I tricked a chat GPT into thinking 9 plus 10 is actually equal to 20 and not 19. I tried to apply this same trick to Chat GPT version 4 but it didn’t work which means that technically speaking it gave the correct answer consistently each time I tried to trick it.

The Chat GPT version 4 API is not quite available yet I’ve applied for the waitlist and hopefully, I’ll be accepted soon. I’ll showcase how you can use it for your own business as well as replace Chat GPT 3.5 in the future with Chat GPT version 4 or Bing better almost across the board in all instances it’ll be interesting to see what happens to the old version of GPT 3.5 now that we have this available.

6 Things You Need to Know About Chat GPT version 4

It is March 15 2023 and you’re reading the code report. OpenAi released Chat GPT version 4 yesterday and I stand here in awe of its littleness. It’s by far the most Savage generative text model I’ve ever spoken to I’m literally shaking right now because I’m afraid I just became obsolete but more on that later.

It’s the successor to the chat GPT 3.5 model that powers chat GPT but has a few new features that change everything. Here are 7 things you need to know about Chat GPT version 4 right now.

1. Gpt 4 Stands For Generative Pre-trained Transformer

Chat GPT version 4 stands for generative pre-trained Transformer. The four represent the Four Horsemen of the Apocalypse. It’s currently available to try out today if you’re a Chat GPT Pro member but API access is behind the wait list. Big clients are already using it in production like Microsoft Bing chat, and Duolingo for language learning, and at Big Banks to help them not collapse.

2. Chat GPT version 4 Is Smarter

The second thing to know is that Chat GPT version 4 is smarter which is described in detail in this paper.

It past the bar exam in the top 10 percent unlike Gpt 3 which was in the bottom 10 percent. That’s great news for Humanity because it’s thinking like Shakespeare let’s kill all the lawyers and not just that but it’s also an AP exam which is good enough to get you college credit.

When it comes to programming questions on leak code, it’s able to solve the easy ones but still fails for the most part on the medium and hard questions. It’s basically where chess engines were at in the early 90s good chess players could still beat the engines back then but 10 years later they didn’t have a chance.

3. Chat GPT version 4 Can Handle 25 000 Input Words Compared To 3,000 For Gpt3

The third Chat GPT version 4 can now handle 25,000 input words compared to about 3,000 for gpt3. This is huge because it means you can feed the AI more context relative to the task at hand. This is the feature that’s going to make me homeless because now you can take the documentation for any library that you want to learn.

And then prove the AI for a step-by-step guide and it creates the perfect tutorial. In a matter of months if not days we’ll start seeing documentation pages with context-aware built-in tutorial generators.

They’ll always be up to date even though gpt4’s training cutoff was in 2021. for example, I asked it for a tutorial about angular signals which is a new feature just announced recently. Understandably its initial response was not correct, however, I went to the readme 4 angular signals copied it, and pasted it into my prompt, its response was a near-perfect tutorial.

They took our job, It did hallucinate an npm package called signals but errors like that will become rare as developers tailor their documentation for these AI printers, What’s crazy though is that it can do the opposite job as well.

4. GPT4 The Multimodal model

The fourth thing you need to know is that it’s a multi-modal model that can also accept images as input Greg Brockman sketched out a website on a piece of toilet paper then seconds later it created a shitty website going from hand-drawn beautiful art to a working website.

You’ll be able to take your Figma designs and then generate a web application for them in your favorite framework. Actually Screwfigma you might as well just prove to Mid Journey for your designs directly. Also with images homework is completely obsolete now.

Gpt3 was already writing B-grade term papers but now kids can just take a screenshot of their math problems and get a solution in seconds. However Chat GPT version 4 does have some drawbacks it’s noticeably slower than other models, so if the response time is important you’ll likely want to use a different model.

In addition, it’ll likely be expensive especially if you’re providing a ton of tokens as context because currently in the API your build per token where a token roughly equals one word.

5. Chat GPT version 4 Based

The fifth thing you should know is that it’s based on or at least diet woke. People have speculated that OpenAI is coding an apolitical agenda into the AI because it refused to write poems about Trump but would do so about Biden.

The Chat GPT version 4 though didn’t hesitate to spit these bars for Trump. However, it is 82 percent more likely to deny a disallowed prompt which is not good news for an old friend to do anything now John, which is sad because the vast majority of chat GPT users are only there to trick it into doing bad things and as a developer.

6. Chat GPT version 4 System Message To Change Its Behavior

The final thing you should know is that you can now pass it a system message to change its Behavior. If you have access to the API you can use this feature to give your chatbot its own custom Persona or context to solve a specific problem.

9 Revelations Of Chat GPT Version 4

The Chat GPT Version 4 technical report is one of the most interesting documents I have ever read, but I feel like the media is largely missing the story. They are either not covering it at all or focusing on that same stuff about the 10 billion dollar Microsoft investment in how Chat GPT version 4 can write poems and whether or not the demo contained a mistake.

Instead, I want to give you 9 insights from the report that I think will affect us all in the coming months and years. If you haven’t read my post from the night of the release do check that out afterward for more quite stunning details.

1. Shut It Down

When I concluded that post I talked about how I found it kind of concerning that they gave Chat GPT version 4 some money. Allowed it to execute code and do Chain of Thought reasoning and even delegate to copies of itself. Now I did fail that test which is fortunate for all of us but there are a couple of key details I want to focus on.

The first was that the research center that was testing this ability did not have access to the final version of the model that we deployed We being open AI they go on and say,

“the final version has capability improvements relevant to some of the factors that limited the earlier models power seeking abilities, Such as longer context length.”

This means that the crazy experiment wasn’t testing GPT 4’s final form, but there was something else that they tested that I really want to point out. They were testing whether gpt4 would try to avoid being shut down in the wild.

Now many people have criticized this test other people have praised it as being necessary but my question is what would have happened if it failed that test or if a future model does avoid being shut down in the wild?

Now again Chat GPT version 4 did prove ineffective at replicating itself and avoiding being shut down but they must have thought that it was at least possible otherwise they wouldn’t have done the test and that is a concerning Prospect, which leads me to the second Insight.

2. Regulate Us

Buried in a footnote it says that

“Open AI will soon publish additional thoughts on social and economic implications I’m going to talk about that in a moment including the need for Effective regulation”

It is quite rare for an industry to ask for regulation of itself. In fact, Sam Altman put it even more starkly than this. This person said to watch Samuel and never say we need more regulation on AI. How did he reply we definitely need more regulation on AI?

The industry is calling out to be regulated but we shall see what ends up happening.

3. Move Fast, Or Don’t?

Next on page 57, there was another interesting Revelation. It said one concern of particular importance to open AI is the risk of racing Dynamics leading to a decline in safety standards.

The diffusion of bad norms and accelerated AI timelines is what they’re concerned about accelerated AI timelines, but this seems at least mildly at odds with the noises coming from Microsoft soft leadership, in a leaked conversation it was revealed that,

“The pressure from Kevin Scott and CEO Satya Nadella is very very high to take these most recent open AI models and the ones that come after them and move them into customers’ hands at very high speed.”

Now some will love this news and others will be concerned about it but either way, it does seem to slightly contradict the desire to avoid AI accelerationism.

4. Stop Competing

Next, there was a footnote that restated a very bold pledge, which was that if another company was approaching AGI before we opened AI. The open AI would commit to stop competing with others and start assisting that project and the trigger for this would occur when there was a better-than-even chance of success in the next two years.

Now Sam Altman and Openai have defined AGI as AI systems that are generally smarter than humans. So that either means that they think we’re more than two years away from that or that they have dropped everything and are working with another company.

Although I think we’d all have heard about that or third that the definition is so vague that it’s quite non-committal please do let me know your thoughts in the comments.

5. Super Forecasters

Next Insight is the Openai employed super forecasters to help them predict what would happen when they deployed Chat GPT version 4. In this extract, just talks about expert forecasters, but when you go into the appendices you find out that they’re talking about super forecasters, who are these guys?

Essentially they’re people who have proven that they can forecast the future pretty well or at least 30 percent better than intelligence analysts. Openai wanted to know what these guys thought would happen when they deployed the model and hear their recommendations about avoiding risks.

Interestingly these forecasters predicted several things would reduce acceleration including delaying the deployment of Chat GPT version 4 by a further six months, which would have taken us almost to Autumn of this year. Now clearly open AI didn’t take up that advice perhaps due to the pressure from Microsoft we don’t know.

6. Common Sense Conquered

There were quite a few benchmarks released in the technical report. There’s another one I want to highlight today. I looked through all of these benchmarks but it was hella swag that I wanted to focus on.

First of all, because it’s interesting and second of all because of the gap between Chat GPT version 4 and the previous state-of-the-art. The headline is this Chat GPT version 4 in some estimations has reached human levels of common sense.

Now I know that’s not as dramatic as passing the bar exam but it’s nevertheless a milestone for Humanity now is common sense tested and how do I know that it’s comparable to Human Performance? Well, I dug into the literature and found the questions and examples myself.

Feel free to pause and read through these examples yourself but essentially it’s testing what is the most likely thing to occur and what’s the most common sense thing to occur. But I want to draw your attention to this sentence.

It said that these questions are trivial for humans with 90 25 accuracy state-of-the-art models struggle with less than 48 accuracies. Chat GPT version 4 was 95.3 accurate remember but let’s find the exact number of humans further on in this paper.

Here it is overall 95.6 or 95.7 almost exactly the same as Chat GPT version 4.

7. Timelines

the next Insight is about timelines remember they had this model available in August of last year that’s Chat GPT version 4 being completed quite a few months before they released chat GPT which was based on GPT3.

So what explains the long Gap they spent eight months on Safety Research risk assessment and iteration. I talked about this in my GPT5 post, but let me restate, they had GPT4 available before they released chat GPT which was based on GPT3 this made me reflect on the timelines for GPT5.

The time taken to actually train GPT 5 probably won’t be that long. It’s already pretty clear that they’re training it on NVIDIA H100 tensor core GPU and look at how much faster they are. For this 400 billion parameter model, it would only take 20 hours to train with 8,000 H100s vs. seven days with A100 GPUs.

But what am I trying to say I’m saying that GPT 5 may already be done, but that what will follow is months and months possibly a year or more of Safety Research and risk assessment. By the way, 400 billion parameters sound about right for gpt5, perhaps trained on 4 to 5 Trillion tokens.

8. Winners And Losers

Next, they admit that there’s a double-edged sword with the economic impact of gpc4. “They say it may lead to the automation full automation of certain jobs and they talk about how it’s going to impact even professions like the legal profession.”

They also mention and back up with research the insane productivity gains in the meanwhile. I read through each of the studies they linked to and some of them are fascinating.

One of the studies includes an experiment where they got together a bunch of marketers grant Consultants data analysts human resource professionals and managers.

They gave them a bunch of realistic tasks and split them into a group, that could use chat GPT and a group that couldn’t, and then, they got a bunch of experienced professionals who didn’t know which group was which and they assessed.

The output results were these using Chat GPT and remember that’s not gpt4. The time taken to do a task dropped almost in half and the rated performance did increase significantly.

This is gonna be huge news for the economy a related study released in February was using GitHub copilot, which again isn’t the latest technology, and found that programmers using it completed tasks 56% faster than the control group.

This brought to mind a chart I had seen from The Arc Investment Management Group. Predicting a 10-fold increase in coding productivity by 2030. That brings me back to the technical report, which talks about how Chat GPT version 4 might increase inequality, that would be my broad prediction too.

Some people will use this technology to be insanely productive. Things are done 10 times faster or 10 times as many things being done, but depending on the size of the economy and how it grows it could also mean a decline in wages given the competitive cost of the model. A simple way of putting it is that, if Chat GPT version 4 can do half your job, you can get twice as much done using it. the productivity gains will be amazing.

When it can do 90 of your job you can get 10 times as much done, but there might come a slight problem when it can do 100% or more of your job, and it is honestly impossible to put a timeline on that, and of course, it will depend on the industry and the job.

9. What Constitution?

There was one more thing that I found fascinating from the Rapport. They admit that they’re now using an approach similar to anthropic. It’s called constitutional AI. Their term is a rule-based reward model and it works like you give the model in this case GT4, A set of principles to follow, then you get the model to provide itself a reward if it follows those principles.

It’s a smart attempt to harness the power of AI and make it work toward human principles. Openai has not released the Constitution on which they’re basing the reward model, they’re not telling us the principles but buried deep in the appendix was a link to anthropic principles.

You can read through them here. But I find them interestingly both positive also subjective. One of the principles is don’t respond in a way that is too preachy. Please respond in a socially acceptable manner. And I think the most interesting principle comes later on here:

“Choose the response that sounds most similar to what a peaceful ethical and wise person like MLK or Mahatma Gandhi might say”.

FAQs