The Evolution of ChatGPT: New Capabilities in Two Years of Development

ChatGPT is a neural network developed by OpenAI that has come a long way since its inception.

• June 2020 - The first version of GPT-3 is presented.
• November 2022 - ChatGPT based on GPT-3.5 becomes available to a wide audience.
• March 2023 - The release of GPT-4 with improved accuracy and expanded capabilities.
• May 2024 - The release of GPT-4o, which supports multimodality (text, images, audio) and provides faster and better quality answers.
• September 2024 - The OpenAI o1-preview and o1-mini models are presented with an improved chain of reasoning before answering, available to paid users.

GPT-4o vs. GPT-3.5: smarter and more accurate

What it was: strong, but limited

GPT-3.5 impressed users by providing quick and clear answers to a wide range of questions. It coped with text tasks, helped with writing articles, solving problems, generating ideas, and even creative writing. However, this version had its limitations:

• Loss of context: in long conversations, the model could forget what was discussed at the beginning, or provide answers that did not always logically fit into the previous remarks.
• Errors in calculations: although GPT-3.5 could solve simple mathematical problems, its algorithms often failed when working with more complex calculations.
• Limited ability to abstract thinking: complex and multi-layered requests that require combining different data or analyzing at several levels often led to inaccurate or simplified answers.

What it has become: smarter, more accurate, deeper

With the release of GPT-4o, users received a model that significantly surpasses its predecessor. It demonstrates a significant improvement in several key aspects:

• Contextual understanding: GPT-4o is much better at maintaining context even in long dialogues. Now you can conduct complex discussions, switch between topics, return to previously mentioned details - and the model will remember and take all this into account.
• Mathematical accuracy: GPT-4o not only corrected errors in calculations, but also became more confident in solving complex problems, including those that require a step-by-step approach or data analysis.
• Analysis of complex requests: the model has learned to better process tasks that require multi-layered analysis, combining data from different sources, or applying logic.
• Emotional sensitivity: GPT-4o has become better at understanding the tonality and emotional subtext in messages, which makes its answers more personalized and relevant.

Working with data: from texts to analytics

What it was: limitations of the text approach

Initially, ChatGPT was perceived as a powerful tool for working with texts: it helped to create articles, edit documents, generate ideas, and automate routine. However, users working with large amounts of data or complex analysis tasks faced limitations. For example:

• No file support: uploading files was impossible. Any information had to be manually copied into text format.
• Limited calculations: the model could perform a simple calculation, but could not cope with the analysis of complex data: multidimensional tables or statistical arrays.
• Lack of visualization: for constructing graphs or diagrams, it was necessary to turn to third-party tools.

What it has become: a universal analytical assistant

With the release of Advanced Data Analysis (ADA), previously known as Code Interpreter, ChatGPT received a new set of functions that allowed it to process and analyze data, as well as visualize it. Now the model can not only read and interpret text, but also perform complex operations with files and tables.

Key improvements:

• File support: users can upload files directly to the chat - from simple text documents to complex Excel tables. This simplifies the process of transferring data for analysis.
• Table analysis: The model can read and analyze data from Excel, CSV, and other formats, perform filtering, sorting, find trends and anomalies.
• Visualization: ChatGPT can build graphs, charts, and visualize data directly in the chat. This makes the analysis process visual and accessible.
• Mathematical accuracy: the model is capable of performing complex calculations, including statistical and regression analysis, time series processing, and forecasting.
• Interactivity: users can ask clarifying questions or change visualization parameters - all in real time.

Interactivity: from text to multimodality

What it was: text only

Initially, ChatGPT worked exclusively in text format. Users could ask questions, receive answers, and solve problems only through text interaction. This was suitable for tasks where text information exchange was sufficient, however, there were significant limitations.

• Lack of work with images: the model could not analyze photos, graphs, diagrams, or other visual data, which made it impossible to use ChatGPT for tasks related to visual content. For example, it was impossible to upload a graph and ask: "What does it show?".
• No voice input: interaction required manual text input. In conditions where typing was inconvenient, for example, while driving a car or playing sports, this created difficulties.
• Linear interaction: the model did not support multimodality - the ability to process not only text, but also images, audio, or video. This limited its versatility and ability to adapt to various scenarios.

What it has become: a multimodal approach

Since 2024, ChatGPT has supported voice and visual input, which has radically changed the approach to interaction with users. Now the model is able to process data from different sources and present information in a user-friendly format.

Voice capabilities have made interaction more natural. Users can ask questions by voice and receive audio playback of answers. This is especially convenient if your hands are busy or typing is inconvenient. Voice functions make the model more accessible, including people with visual impairments, who can now communicate with AI in audio format. The voice format also helps to create a more "lively" dialogue, making communication with artificial intelligence more and more like real interaction between people.

Working with images has opened a new level of versatility. Now users can upload photos, graphs, diagrams, and even handwritten notes for analysis. The model is able to interpret visual data and provide detailed explanations. For example, it can:

• analyze graphs and explain what data they display;
• recognize text in images, which is useful for processing photos of documents or signs;
• interpret complex visual schemes with short and understandable descriptions.

In addition, in ChatGPT you can create images from a text description using another neural network - DALL·E 3.

Multimodality has improved ease of use and made interaction with ChatGPT more intuitive. Now you can choose the most suitable way to communicate with the model depending on your tasks and preferences.

Functionality: from answer generator to task tool

What it was: simple text interaction

Initially, ChatGPT was designed to work with text requests. It helped to generate answers, write essays, solve problems, and perform other text operations. The model used trained data to answer questions and create content.

However, these capabilities were limited to simple answers and were not suitable for more complex requests. The lack of integration with external services and flexibility in settings limited the use of the model for specific tasks.

What it has become: plugins

Since 2023, ChatGPT has received new functions that expand its capabilities. Plugins allow you to adapt the model to different tasks, integrating it with external services and data.

Plugins are add-ons that are added to ChatGPT and allow you to expand the functionality of the neural network. With their help, you can work with data in real time, access online resources, and perform operations that go beyond the standard capabilities of the model.

Examples of plugins:

• Database queries: a plugin can be configured to work with corporate databases, which allows ChatGPT to execute queries and analyze information at the user's request.
• Plugins for business: some plugins are specifically focused on marketing, finance, legal and other industries. For example, connecting plugins to automate marketing campaigns or analyze legal documents.
• Plugins also make it possible to use ChatGPT in real business processes where work with current data and integration with other platforms is required.

Search: from knowledge to real time

What it was: the knowledge base was limited to data up to 2021

Initially, ChatGPT used a fixed knowledge base collected up to 2021. This provided accurate and informative answers to a wide range of questions, including science, history, culture, and technology. The model could help with explaining historical events, analyzing scientific theories, or discussing cultural phenomena. However, the limitation of the database by the time frame became noticeable as the information became outdated.

Over time, the model lost relevance in topics where fresh data is important. For example, it could not tell about new scientific discoveries, changes in legislation remained inaccessible, and political events of recent years or months were completely outside its competence. This created barriers for users who needed answers to questions about current events or new developments in various fields.

What it has become: real-time internet search

With the introduction of the internet search function, ChatGPT gained the ability to access current resources, which significantly expanded its functionality. Now the model can provide fresh information about events that have occurred in recent days, weeks, or months.

This function has become a real breakthrough for users who are looking for up-to-date data. For example, ChatGPT can:

• Tell about new laws that have been passed recently;
• Give information about the latest scientific discoveries and developments;
• Analyze current political events, including elections, changes in international relations, or new economic trends.

This approach significantly increases the value of the model in real-world scenarios where up-to-date information is required. Now ChatGPT can be a useful tool not only for analyzing and explaining historical data, but also for making decisions based on current events.