# Practical uses of artificial intelligence

> What can artificial intelligence offer in web and app development, and how can we put it to work? And what do the singularity, AI hype and AGI mean?

In December 2022, OpenAI released ChatGPT to the public and took us into a new era, one that is changing technology, the economy and society. Neural networks and LLMs (large language models) are nothing new; they have been around in academia and research for decades. But the launch of ChatGPT, and now GPT-4, took everyone outside those fields by surprise. In fact, even its creators are surprised every day by what this new technology can do.

![The family tree of large language models (LLMs)](https://webikon.sk/cdn-cgi/image/width=2048,format=auto,fit=scale-down/https://media.webikon.sk/originals/db/c7/dbc7c49ce114300f.png)

### AI hype, AGI and the singularity

Before I get to LLMs, and ChatGPT in particular, I'd like to sum up the current trends and the state of artificial intelligence in general. We live in an age of social media, short dramatic episodes and the usually unreasonable hyperbole that comes with them. When it comes to AI, it is becoming quite clear that, for all the sensational news, clickbait and overestimation, we are all entering unknown territory. Growth is exponential, technological progress keeps accelerating, and one thing Stephen Hawking warned us about is building self-improving AI. That is essentially what most people interested in AI are doing now. Terms like “AGI” and “the singularity” have suddenly moved from science fiction into reality. For decades, Ray Kurzweil has predicted that AI will reach human level around 2029 and that the singularity will arrive in 2045, while other researchers and scientists assumed it would take much longer, or that it could not be achieved at all. Yet with every decade, and especially in recent years, the general consensus in the scientific community has moved much closer to Kurzweil's predictions.

### So what is the singularity?

I have just watched [this TED talk by Ray Kurzweil himself](https://www.youtube.com/watch?v=PVXQUItNEDQ), recorded in 2014. Not only did he anticipate, with surprising accuracy, the direction taken by technologies like ChatGPT, Bing and Bard that we are experiencing right now, he also predicted advances in nanobots that **we can implant in our bodies** for various purposes. One of them is the ability of nanobots to connect to the cloud and draw on computing resources and information beyond what the brain has today. In other words, your mind would be connected directly to additional sources of information and computing power, so you would **know more and think faster**.

Kurzweil uses **the singularity** for the point at which the human mind merges with artificial intelligence and technological progress speeds up beyond our ability to predict it. One of the consequences he describes is the chance to **separate our minds from our bodies** and so achieve immortality. While his sense of where language models were heading proved surprisingly accurate, I can't judge how far along we are with nanobots. But the progress is clear, and the path Ray Kurzweil outlined years ago seems to be merging more and more with our everyday lives.

### Artificial General Intelligence (AGI)

AGI stands for **Artificial General Intelligence**, in other words **intelligence that is general rather than specialised**. I first heard the term in December, listening to an interview with Sam Altman (CEO of OpenAI). Fortunately, it is a little easier to grasp than the singularity described above.

![Artificial intelligence and the brain](https://webikon.sk/cdn-cgi/image/width=445,format=auto,fit=scale-down/https://media.webikon.sk/originals/e1/60/e1608ecfaf3fbdda.jpg)

Today we have narrow AI systems: intelligences that can work within a specific domain or with specific tools. For example, AI that can identify cats in pictures, or recommend films on whichever streaming service you use. These have developed rapidly in recent years, but they are still essentially narrow systems. AGI, then, should be something that **goes beyond a single domain**: a general, all-knowing intelligence that **understands everything about society, the world and the universe**. And we are on that path too, especially now with current LLM technology.

### What AI can do today

Let's move from philosophy and hyperbole to a more practical overview of where **today's artificial intelligence** stands, what it **can really do and what tools are available**.

The first example is the very realistic [deepfake video of Tom Cruise](https://www.youtube.com/watch?v=iyiOVUbsPcM) that appeared about **two or three years ago**. The developers managed to create a photorealistic digital avatar of Tom Cruise from footage of him, **without his involvement**. The avatar was worked into the video without a hitch, following a scripted dialogue. The example shows that you can create a digital likeness of anyone and manipulate it into doing anything. Rendering Tom Cruise took a lot of time and resources at first, but over the past few years the technology has advanced so far that convincing digital replicas can be made **with far fewer resources** and much faster.

Another example is the **real-time animation of a MetaHuman model**. [This demo](https://www.youtube.com/watch?v=pnaKyc3mQVk&t=68s) took place a few months ago and was presented by the makers of **Unreal Engine**, which is widely used in video games. During the presentation, an actress recorded a short 10-second video on her mobile phone, showing a range of emotions. The video was then fed into AI software that generated a digital replica of the actress on screen. The digital avatar copied her movements and expressions flawlessly. What's more, the facial expressions could be transferred to a different digital model, creating a **realistic and convincing representation of a person** who never actually performed those actions.

![A demo of MetaHuman technology](https://webikon.sk/cdn-cgi/image/width=2046,format=auto,fit=scale-down/https://media.webikon.sk/originals/37/83/3783988735f8f4f1.png)

![A real-time demo of MetaHuman technology](https://webikon.sk/cdn-cgi/image/width=2032,format=auto,fit=scale-down/https://media.webikon.sk/originals/f6/b8/f6b86b4eb7787d22.png)

On top of that, there are now AI systems that can alter a person's appearance in real time during a live broadcast. Photoshopped images have been common for some time, but these AI systems can **change a person's appearance and voice in real time**, making it almost impossible to tell real content from digitally altered content.

The last example is the concept of **Alpha Persuade**, discussed [in the presentation “The AI Dilemma”](https://www.youtube.com/watch?v=xoVJKj8lcNQ). Alpha Persuade combines two elements: **the ability to replicate a person's voice** from a five-second recording, and the power of AlphaGo, the artificial intelligence developed by Google that mastered the game of Go by simulating millions of games in a very short time. By simulating a conversation in the voice of someone close to you, Alpha Persuade could convincingly manipulate people into doing things they would normally consider very unlikely. Alpha Persuade is not currently available to the public, but the theoretical possibility of using it already exists today.

### AI tools and development

On YouTube, on Google or even through Bard and Bing, you will find plenty of other AI tools that can do surprising things. They are built for different fields and work with **video**, **audio**, **presentations**, **social media**: basically any digital content.

One popular platform **for developers** is **[Hugging Face](https://huggingface.co/)**, with a large number of AI models built for specific purposes. Hugging Face gives you access to these models, and you can also use the GPT API (from OpenAI) to build applications. ChatGPT still has certain advantages, but in recent months open-source models have appeared that work in quite a similar way. It shows how fast the whole field is moving.

![The Hugging Face platform](https://webikon.sk/cdn-cgi/image/width=1218,format=auto,fit=scale-down/https://media.webikon.sk/originals/8e/94/8e945f29a2035cde.png)

#### How GPT differs from ChatGPT

**GPT** stands for **“Generative Pre-trained Transformer”**. It is the underlying technology that large language models are built on.

**ChatGPT**, on the other hand, is a specific GPT model that has been **trained on conversations** and on instructions for completing text prompts.

Anyone who develops applications or works with language models uses the **GPT API** and the base GPT model. It can be fine-tuned or trained on specific data to suit the user's needs. As a result, we can already see several ways of offering AI-based products and services that are available for free or for a very reasonable fee.

### AI products and services

Many agencies and companies may explore artificial intelligence in all its forms. **At Webikon**, we want to focus mainly on LLMs, the text-based generative models such as GPT, and gradually make use of their ability to work with images and audio as well (speech-to-text and text-to-speech).

#### Intelligent search

First of all, GPT makes it easy for us to build **intelligent search** into any website. We prepare the content and, using the “embeddings” offered by the GPT API, we can **translate the content into vectors** that carry real semantic meaning.

When a visitor uses the website's search, we take their query, translate it into a vector as well, and compare it with the vectors stored in the database. The result is **semantic search based on meaning** rather than old-fashioned, simple keyword matching. The search engine genuinely understands what the user is asking for and offers them matching results.

![Intelligent search](https://webikon.sk/cdn-cgi/image/width=1194,format=auto,fit=scale-down/https://media.webikon.sk/originals/45/f2/45f2c510d5bdebfe.png)

#### Intelligent chatbot

GPT also lets us offer a **chatbot on your website**. All it takes is extending the first example, intelligent search. The website's data is already converted into vectors; we convert the user's input into vectors too and compare them. Then we can simply switch on context memory, and you can hold a fluent conversation with the system. That gives us an intelligent chatbot that knows everything about your website and can **talk to you about it**.

#### Intelligent assistant

Going even further, we can build our own internal intelligent assistant that **increases our productivity** and gives us **objective feedback**, as well as new options for our processes and projects.

Again, this builds on the previous two examples, with a more complex data structure and prompt engineering techniques, so that the assistant has the knowledge it needs and answers to the point. It might be more accurate to call it an agent: with GPT-4, these models are approaching “reasoning” capabilities, and we can connect these reasoning language models to the internet and let them run a script or code through a plugin. So we are moving beyond a passive listener that simply waits for your input, towards an **active agent** that actually thinks and acts.

Finally, we are currently exploring ways of **bringing images and audio into our processes**, to build our own voice-controlled AI agent and speed up prototyping and app design. The main goal is to deliver the best quality much faster, and today's AI gives us everything we need to **reach that goal**.

![Rostislav Kavan, AI Developer](https://webikon.sk/cdn-cgi/image/width=552,format=auto,fit=scale-down/https://media.webikon.sk/originals/b3/46/b3460d25cd31ab27.png)

AI Developer

#### Rostislav Kavan

I have always enjoyed history, philosophy and languages. At secondary school I added programming languages to the list. What I like about programming is that it keeps pushing you forward; there is always something to improve. That is also why ChatGPT, as a large language model, became my new hobby practically overnight. I have also set up a [Facebook group](https://www.facebook.com/groups/1194802001431491/) for ChatGPT enthusiasts, where we share news from the world of artificial intelligence. You are welcome to join us too.

### Let's work out where AI fits in your project.

We'll go through your processes and data and tell you what can be automated today and what doesn't make sense yet. Even if the answer is that you don't need AI for now.

[Discuss AI for your business](/en/contact/)

[How we work with AI](/en/services/ai-solutions/)

---

Source: https://webikon.sk/en/blog/artificial-intelligence/