How not to get lost in innovations: a guide from AI market expert Gleb Dobzhansky

New products in the world of generative AI appear by the dozens every week. However, the leaders of the race and the direction of movement are already known. Hlib Dobzhansky, Vice President of Master of Code Global, a company that develops AI-based products, explains how entrepreneurs can navigate the AI universe

Everything changes almost every day in the field of generative AI. These are tectonic changes. A technical revolution, if you will.

ChatGPT successfully passes the US medical license exam. An artificial intelligence robot wins a famous photo contest. Morgan Stanley tests a chatbot from OpenAI for its 16,000 financial advisors.

Remember how the first computers were distributed. In 1977, Ken Olsen, founder and CEO of Digital Equipment Corporation, said there was no reason why anyone would want a computer in their home. It was a mistaken assessment of the potential.

Now we're going through the same revolution - the transition from a graphical interface to a conversational one. Everyone will have a personal assistant in the form of an artificial intelligence chatbot.

Microsoft and OpenAI: why did they succeed, unlike Google or Meta?

All major tech companies have been developing AI. It was OpenAI that became a pioneer in the industry and set its own standards in generative artificial intelligence. Other companies missed this leap, although they have a chance to catch up. Some use dirty methods, calling on the world to pause AI development while developing their own similar AI products.

What is the secret of OpenAI's success?

• ChatGPT is a good example of what AI can do. Its interface is clear and contains many tips and ideas that help you start using it without additional training.
• ChatGPT is available to users and has been released in Public Beta in many countries quite quickly.
• Microsoft (the largest investor in OpenAI) has been successful in selling services based on OpenAI: Copilot, ChatGPT Plus paid. OpenAI is also sold as part of Microsoft's cloud services and is used in Salesforce CRM, an integrated customer relationship management platform, as an Einstein GPT assistant Salesforce Announces Einstein GPT.
  A sophisticated moderation system that ensures that the bot responds correctly.
• Text generation is the tip of the iceberg

Information retrieval, writing program code, image generation, building personal assistants, and expert systems already exist on the basis of generative artificial intelligence.

According to Gartner, while AI-generated data accounted for less than 1% of all data until recently, by 2025 this figure will increase to 10%.

Search.

ChatGPT works in Microsoft's Bing search service and provides alternative search results. Over the previous month, Microsoft finally managed to break the 100 million mark of active users per day. Downloads of the Bing app jumped eightfold after the integration of artificial intelligence.

ChatGPT cannot replace search engines. The biggest problem is data relevance. Its base is the information available at the time of its training. Events that occur in real time are not available to it. Fact verification also suffers.

Even with these limitations, ChatGPT and Bing are becoming a growing threat to Google, which has a market share of over 80%. Therefore, everyone is waiting for the company's response.

Writing program code

The GPT-3 model was trained on data from the largest open source software repository GitHub. This allowed the chatbot and products that use it to understand and create new code on demand.

Microsoft has developed and sells a GitHub Copilot plugin that helps to write code, tests, or documentation by understanding the context. It's similar to pair programming, where your imaginary colleague can take away all the routine work or offer some alternative solution.

Image generation

At first, AI image generation looked like interesting research, but with the latest updates, Midjourney creates pictures that are hard to distinguish from those created by humans.

Adobe has announced the ability to create AI images in Photoshop. A model trained on the company's stock image database can create a new image based on a description or edit an existing one (for example, change the background, add a hat or glasses).

These tools are being actively integrated by the creative sector. According to a survey of 1000 creatives in the US, 71% will use AI tools in their professional work. Among UI/UX designers, the figure is 91%.

Expert systems

For expert systems, even such large models may not be enough to find the necessary facts or require a high speed of data generation. That's why specialized LLM models appear.

The BloombergGPT model was recently announced, with a training focus on financial data. The answers of this model can already be perceived as an expert system that will help make more informed investment decisions, reduce the risk of financial losses, implement more efficient trading strategies, and prevent fraud.

Against this backdrop, it is quite funny to recall how the chatbot was ridiculed on social media for making mistakes and inventing. While some people were looking for the areas where AI does not work, others were looking for the areas where it does.

The task of every business professional is to think about where AI can be applied. It is not about experiments or an additional source of income for the business. It is a matter of business survival.