13 Breakthrough Events in Generative AI in 2023

The year 2023 has proven to be one of significant breakthroughs in the field of generative artificial intelligence (GenAI), and the related advances have become revolutionary. In this article, we present our subjective list of the most critical events in the area of GenAI in the last year that had the most significant impact on the development of this field and our lives.

The 2023 was full of the expansion of large language models and various tools based on them. What accelerated the process and made this breakthrough significant was not only the quality of the solutions being developed but also their democratisation. Open-source tools have achieved unprecedented levels of development, changing the technological landscape and the face of most industries. The widespread availability of GenAI tools unlocks unmatched opportunities for innovation and growth for individuals and businesses.

Crucial breakthroughs in GenAI in 2023 [subjective list]

January

1. ChatGPT reached a user base of 100 million

Launched by OpenAI and backed by Microsoft with an investment of approximately 10 billion, ChatGPT became an internet sensation and was considered synonymous with the breakthrough in the field of GenAI. It reached a user base of 100 million in just two months after publication, and the first million in just five days. It is the only application that has recorded such a rapid pace of development.

How long it took other start-ups to reach their first million users?

• Twitter – 24 months
• Facebook – 10 months
• Dropbox – 7 months
• Spotify – 5 months
• Instagram – 3 months

ChatGPT is a powerful tool for automation, analysis, process optimisation and decision-making, revolutionising business, education and everyday life. Its role as a tool supporting creativity and work efficiency is undeniable.

Potential use cases include:

• customer service and support;
• diagnostics and Healthcare;
• education and training;
• marketing and advertising.

More about the benefits of ChatGPT: ChatGPT and GenAI – why NOW is the time to understand them?.

February

2. Google launched its Bard

The release of ChatGPT challenged Google, which holds a vast competitive advantage on the Internet. The emerging competition between Google Bard and ChatGPT has ushered in one of the most exciting chapters in the history of AI development, driving innovation in the technology industry.

At launch, Bard demonstrated similar features to ChatGPT, such as:

• generating various text formats (poems, code, scripts, music, e-mails, lists, etc.);
• real-time language translation;
• answering questions in a comprehensive and informative way;
• access information from the real world and process it using the Google search engine.

3. Microsoft has integrated ChatGPT with the Bing search engine

Unlike traditional query search tools, Bing was intended to act as an interactive chatbot, allowing users to have deeper and broader conversations for specific search needs.

While ChatGPT's and Bard's bots are suitable for everyday use by employees, they were not recommended for large companies at the time because they used anonymous data from user interactions to improve model performance and capabilities. Microsoft's solution in the form of "Bing Chat Enterprise", over time, gave companies solid security features and enabled them to take full advantage.

Bing Chat introduced features such as:

• adding new visual elements;
• option for users to save and export chat conversations;
• improved browser appearance and summary capabilities in Edge;
• integration of plug-ins that facilitate interactions with third parties.

March

4. Announced GPT-4 – the latest version of the language model that supports ChatGPT

GPT-4 stood out for its extraordinary proficiency in document analysis, creative writing, translation, coding, complex problem solving, and, above all, its image recognition function. At the time of its announcement, it boasted 280 billion ML parameters, and although it was already impressive then, it now has over 1.7 trillion of them.

Companies that gained early access to GPT-4 began implementing it in areas such as financial analysis, payment processing, and language learning. It started to be used for everyday tasks such as transcription, summarising meetings and e-mails. And GPT-4's image interpretation capabilities allow you to take spreadsheets, PDF files, or handwritten notes and turn them into presentations or websites. OpenAi announced accelerating knowledge-based work such as data visualisation, copywriting, translation, editing and creative image generation.

July

5. Meta publishes the open-source Llama 2 model

The release of Meta Llama 2's as “open source” sparked a controversial debate about the definition of open source in AI. Although hailed as a significant contribution, the limitations of Llama 2's scale raised questions about its genuine openness. It initiated discussions about the need to redefine licensing.

6. Anthropic makes the Claude chatbot publicly available in the US and GB

With the release of Claude 2, Anthropic made its chatbot available to users first in the US and UK and gradually in other countries. They did this four months after its initial announcement in March when they released an alpha test version for business.

Claude 2 is similar to ChatGPT from OpenAI and Bard from Google in terms of translating text, writing code and answering various questions. However, according to Anthropic, it is safer and more reliable because it follows a set of rules that allow it to self-correct responses rather than relying on human moderators. Claude can also respond to extended suggestions, making it suitable for viewing longer business or legal documents.

August

7. Amazon invests up to $4 billion in start-up Anthropic (creators of Claude)

The deal included a $1.25 billion initial transaction that gave Amazon a minority stake in the company. It may seem minor compared to the $13 billion that Microsoft has invested in OpenAI until now. Still, it is a clear signal that Amazon is participating in an intensifying AI arms race.

As a result of this merger:

• Amazon Web Services (AWS) becomes the primary cloud service provider for Anthropic;
• Anthropic will leverage AWS's Trainium and Inferential chips to build, train, and deploy Claude and future core models;
• AWS gains exclusive, early access to Anthropic technology.

AWS users – from large corporations to start-ups to individual developers – can take advantage of Claude and other Anthropic models through the Amazon Bedrock platform. It allows you to build advanced AI applications using foundation models (FMs) that can significantly transform how companies undertake and implement business tasks.

September

8. New ChatGPT voice and image features announced

The new features enable voice conversations with the chatbot or sharing images instead of relying solely on typed suggestions. It was initially made available to paid subscribers.

Using ChatGPT's voice capabilities includes, for example:

• virtual assistant – responds to voice commands, schedules meetings, and provides information in real-time;
• practising a new language – you receive feedback and improve your pronunciation;
• benefiting from voice brainstorming by content creators, making the process more dynamic and efficient.

At the time of the announcement, it was noted that while ChatGPT's ability to create synthetic voices and images offers excellent creative opportunities, it also carries potential dangers, such as impersonation and fraud.

October

9. Microsoft has made Microsoft 365 Copilot available as part of a free Windows 11 update

The release to a broader group of users followed the earlier announcement of the tool in March and early access for approximately 600 companies in May. Now Microsoft 365 Copilot is also available on Bing, which has added support for the latest DALL•E 3 (trained on 12 billion images) and an update to Bing Chat Enterprise, making it more mobile and visual, and on Edge, with the announcement for corporate clients with the Microsoft 365 Chat service.

With over 150 new features, the Windows 11 update becomes one of the most ambitious yet, bringing new AI-powered experiences in apps like Paint, Photos, Clipchamp and more directly to your Windows PC. Microsoft 365 Copilot is an AI assistant that has completely changed the character of our work.

Unique features include:

• integration with everyday work tools, e.g. Teams, Excel, Word, Outlook, PowerPoint;
• users can perform tasks such as writing a blog, creating presentations;
• updating calendars and summarising them faster and more effectively;
• Microsoft Search to search and respond to information as prompted.

November

10. Release of GPT-4 Turbo, the latest version of OpenAI

Faster, more accurate and more efficient model of GPT-4. GPT-4 Turbo's advancement relies on knowledge of world events until April 2023. Its 128k context window allows it to process the equivalent of over 300 pages of text in a single query. The features of GPT-4 Turbo, as well as AI solutions that tailor chat to individual needs, make it an ideal solution for companies requiring fast and reliable AI solutions.

Its ability to quickly process large amounts of data is a game changer in the business environment by:

• faster business decision-making;
• improved customer interactions;
• improved data analysis capabilities.

At the same time, the Assistants API was introduced, which has all the power of the core GPT models and some fantastic features like integrating your own data, executing Python code, and using custom functions and external APIs. It was created to help developers and application developers integrate AI features into their applications and overcome the limitations of the context window. The innovation of Assistants API lies in the ability to create a specialised “assistant” that can follow precise instructions and access external knowledge bases. The engine supporting the Assistants API is Code Interpreter, a proprietary tool developed by OpenAI.

Developers can also:

• integrate DALL•E 3 directly into your applications and products via the image API;
• generate natural speech from text using the text-to-speech API – this is a great revolution that makes it easier to create and generate images and, at the same time, allows for voice summarisation of created products.

December

11. The EU announced the world's first AI legislation – the “AI Act”

The next step to adopting the AI Act will be voting in the European Parliament committees – IMCO and LIBE and then voting on the regulation at the plenary session of all MEPs in March or April 2024.

From the beginning, and dating back to 2021, this project has been accompanied by strong emotions and stormy conversations. At that time, the draft AI Act did not yet include generative and essential AI systems – currently the most important ones. The subject of disputes between representatives of EU institutions was which category the AI applications that raise the most significant concerns would be placed in.

The AI Act divides AI applications into four categories and imposes specific obligations depending on them.

• unacceptable solutions – will be prohibited;
• high-risk applications – will be approved for use under strict control;
• limited risk applications – will be covered by information obligations;
• minimal risk AI applications – they will avoid regulation.

According to Euroactiv, France, Germany, and Italy have asked for a more lenient regulatory system for AI models with wide applications (such as Chat GPT and Bard). They argued that they wanted to avoid clipping the wings of European start-ups that could create solutions in this field that would be competitive with American ones.

12. Google presented the Gemini model (former Bard)

Gemini's creator, Google, claims it is the largest and most potent AI model. It is a direct competitor to existing models and has five times the computing power of GPT-4. Made available to the public in early December, it aroused much controversy because, as it later turned out, Google had faked parts of the Gemini AI demo. From the technological side, it is a new, advanced language model based on Tensor Processing Units v4 and v5e technology.

It has 3 versions:

• Gemini Pro is a versatile general-purpose AI model. It is an ideal solution for companies and developers who need a powerful but less specialised AI tool.
• Gemini Ultra is designed for the most demanding and complex AI applications. It is ideal for large research projects, advanced data analysis, and complex problems requiring deep learning and reasoning.
• Gemini Nano is a solution for mobile devices and mobile applications such as Pixel 8 Pro and is optimised for performance and power consumption.

Firstly released in the US and with time in Asia-Pacific, Latin America, North America and Africa. It was not accessible in Europe from the beginning due to problems with ongoing work on EU law. For now (May’24) it is already possible to use Gemini in UE.

13. The New York Times filed a lawsuit against OpenAI and Microsoft

The claim concerns a violation of the principles of fair competition. According to the Daily's representatives, AI first learned from thousands of texts published in the NYT, and now it creates competitive texts in which the influence of the style of the newspaper's journalists can be felt.

According to the lawsuit, Microsoft and OpenAI are making vast amounts of money from this. NYT feels threatened by losing “billions of dollars” in potential future profits and unfair competition from services like Copilot. The journal demands that LLM models stop using its study materials. The BBC, CNN and Reuters also have a similar opinion and blocked the so-called crawlers, combing the web for texts that Copilot/ChatGPT use to create their articles.

Conclusions and forecasts

GenAI is becoming a powerful force in shaping the future – from transforming business strategies and operating models to redefining artistic creativity and accelerating scientific discovery. It is also clear that the possibilities offered by this technology are both enormous and complex. The revolution associated with it is not only technological prowess but also paving the way for a smarter, more creative and more efficient future.

Summary

As organisations explore and integrate GenAI technologies, the focus in 2024 will be even stronger on risk management and further legal aspects pioneered in the EU to ensure that the power and potential of AI are used responsibly and effectively for the benefit of all humanity.

If you want to stay up to date with the next breakthroughs in the field of AI, be sure to subscribe to our channel, Beyond AI – YouTube. There, we discuss the latest discoveries in the area of GenAI and will also be happy to be your guide in this dynamic field. See You soon!